Generative adversarial networks (GANs) have demonstrated significant progress in unpaired image-to-image translation in recent years for several applications. CycleGAN was the first to lead the way, although it was restricted to a pair of domains. StarGAN overcame this constraint by tackling image-to-image translation across various domains, although it was not able to map in-depth low-level style changes for these domains. Style mapping via reference-guided image synthesis has been made possible by the innovations of StarGANv2 and StyleGAN. However, these models do not maintain individuality and need an extra reference image in addition to the input. Our study aims to translate racial traits by means of multi-domain image-to-image translation. We present RaceGAN, a novel framework capable of mapping style codes over several domains during racial attribute translation while maintaining individuality and high level semantics without relying on a reference image. RaceGAN outperforms other models in translating racial features (i.e., Asian, White, and Black) when tested on Chicago Face Dataset. We also give quantitative findings utilizing InceptionReNetv2-based classification to demonstrate the effectiveness of our racial translation. Moreover, we investigate how well the model partitions the latent space into distinct clusters of faces for each ethnic group.

Skyline plays a pivotal role in mountainous visual geo-localization and localization/navigation of planetary rovers/UAVs and virtual/augmented reality applications. We present a novel mountainous skyline detection approach where we adapt a shallow learning approach to learn a set of filters to discriminate between edges belonging to sky-mountain boundary and others coming from different regions. Unlike earlier approaches, which either rely on extraction of explicit feature descriptors and their classification, or fine-tuning general scene parsing deep networks for sky segmentation, our approach learns linear filters based on local structure analysis. At test time, for every candidate edge pixel, a single filter is chosen from the set of learned filters based on pixel's structure tensor, and then applied to the patch around it. We then employ dynamic programming to solve the shortest path problem for the resultant multistage graph to get the sky-mountain boundary. The proposed approach is computationally faster than earlier methods while providing comparable performance and is more suitable for resource constrained platforms e.g., mobile devices, planetary rovers and UAVs. We compare our proposed approach against earlier skyline detection methods using four different data sets. Our code is available at https://github.com/TouqeerAhmad/skyline_detection.

Large–scale visual sensor networks have become progressively an essential part of our daily lives underpinning many technological, financial, and social advancements today, with applications in smart cities, traffic monitoring, environmental pollution control, public safety, and crime prevention.

Craters are among the most abundant features on the surface of many planets with great importance for planetary scientists. They reveal chronology information about planets and may be used for autonomous spacecraft navigation and landing. Although numerous research efforts have been carried out in the field of crater detection, existing crater detection algorithms (CDAs) are only helpful in a limited number of applications. A promising crater detection approach involves two main steps: 1) hypothesis generation (HG) and 2) hypothesis verification (HV). During HG, potential crater locations are detected. The validity of the hypothesized crater locations is then tested in a HV step. In this context, we discuss some commonly used algorithms for HG such as highlight-shadow region detection and Hough transform as well as our novel and enhanced algorithms based on interest point detection and convex grouping. A key objective of this paper is to analyze their performance while paying special attention to how they affect the accuracy of the verification step. To deal with different size craters, we focus on multiscale HG. For HV, we have chosen convolutional neural networks which have recently achieved state-of-the-art performance in many computer vision applications. Due to the variation of test sets in the literature, it is often challenging to compare the performance of different CDAs in a fair way. In this paper, we present a comprehensive performance evaluation and comparison of CDAs. Each algorithm has been trained/tested using common data sets generated by a systematic approach.

Horizon or skyline detection plays a vital role towards mountainous visual geo-localization, however most of the recently proposed visual geo-localization approaches rely on user-in-the-loop skyline detection methods. Detecting such a segmenting boundary fully autonomously would definitely be a step forward for these localization approaches. This paper provides a quantitative comparison of four such methods for autonomous horizon/sky line detection on an extensive data set. Specifically, we provide the comparison between four recently proposed segmentation methods; one explicitly targeting the problem of horizon detection[2], second focused on visual geo-localization but relying on accurate detection of skyline [15] and other two proposed for general semantic segmentation - Fully Convolutional Networks (FCN) [21] and SegNet[22]. Each of the first two methods is trained on a common training set [11] comprised of about 200 images while models for the third and fourth method are fine tuned for sky segmentation problem through transfer learning using the same data set. Each of the method is tested on an extensive test set (about 3K images) covering various challenging geographical, weather, illumination and seasonal conditions. We report average accuracy and average absolute pixel error for each of the presented formulation.

This paper investigates the use of Convolutional Neural Networks for classification of painted symbolic road markings. Previous work on road marking recognition is mostly based on either template matching or on classical feature extraction followed by classifier training which is not always effective and based on feature engineering. However, with the rise of deep neural networks and their success in ADAS systems, it is natural to investigate the suitability of CNN for road marking recognition. Unlike others, our focus is solely on road marking recognition and not detection; which has been extensively explored and conventionally based on MSER feature extraction of the IPM images. We train five different CNN architectures with variable number of convolution/max-pooling and fully connected layers, and different resolution of road mark patches. We use a publicly available road marking data set and incorporate data augmentation to enhance the size of this data set which is required for training deep nets. The augmented data set is randomly partitioned in 70% and 30% for training and testing. The best CNN network results in an average recognition rate of 99.05% for 10 classes of road markings on the test set.

Automatic crater detection in planetary images is an important task with many applications in planetary science, spacecraft navigation, landing, and control. Typically, crater detection algorithms consist of two main steps: candidate crater region extraction and crater verification. Various methods have been proposed for extracting candidate crater regions, ranging from detecting circular/elliptical regions to detecting highlight and shadow regions. For crater verification, powerful feature extraction and machine learning techniques have been employed. While this two-step approach can be efficient and robust, inaccuracies in the candidate crater region extraction step can result in mislocalized crater regions which could affect verification performance. In this paper, we investigate the robustness of various feature extraction methods to mislocalized crater regions. Using features which are robust to localization errors but also choosing a more representative training set has yielded significant performance improvements on an extensive dataset from the Lunar Reconnaissance Orbiter (LRO).

Horizon line is a promising visual cue which can be exploited for robot localization or visual geo-localization. Prominent approaches to horizon line detection rely on edge detection as a pre-processing step which is inherently a non-stable approach due to parameter choices and underlying assumptions. We present a novel horizon line detection approach which uses machine learning and Dynamic Programming (DP) to extract the horizon line from a classification map instead of an edge map. The key idea is assigning a classification score to each pixel, which can be interpreted as the likelihood of the pixel belonging to the horizon line, and representing the classification map as a multi-stage graph. Using DP, the horizon line can be extracted by finding the path that maximizes the sum of classification scores. In contrast to edge maps which are typically binary (edge vs no-edge) and contain gaps, classification maps are continuous and contain no gaps, yielding significantly better solutions. Using classification maps instead of edge maps allows for removing certain assumptions such as the horizon is close to the top of the image or that the horizon forms a straight line. The purpose of these assumptions is to bias the DP solution but they fail to produce good results when they are not valid. We demonstrate our approach on three different data sets and provide comparisons with a traditional approach based on edge maps. Although our training set is comprised of a very small number of images from the same location, our results illustrate that our method generalizes well to images acquired under different conditions and geographical locations.

This paper introduces an advanced rover localization system suitable for autonomous planetary exploration in the absence of Global Positioning System (GPS) infrastructure. Given an existing terrain map (image and elevation) obtained from satellite imagery and the images provided by the rover stereo camera system, the proposed method determines the best rover location through visual odometry, 3D terrain and horizon matching. The system is tested on data retrieved from a 3 km traverse of the Basalt Hills quarry in California where the GPS track is used as ground truth. Experimental results show the system presented here reduces by over 60% the localization error obtained by wheel odometry.

Due to the availability of easy-to-use and powerful image editing tools, the authentication of digital images cannot be taken for granted and it gives rise to non-intrusive forgery detection problem because all imaging devices do not embed watermark. We investigated the detection of copy-move and splicing, the two harmful types of image forgery, using textural properties of images. Tampering distorts the texture micro-patterns in an image and texture descriptors can be employed to detect tampering. We did comparative study to examine the effect of two state-of-the-art best texture descriptors: Multiscale Local Binary Pattern (Multi-LBP) and Multiscale Weber Law Descriptor (Multi-WLD). Multiscale texture descriptors extracted from the chrominance components of an image are passed to Support Vector Machine (SVM) to identify it as authentic or forged. The performance comparison reveals that Multi-WLD performs better than Multi-LBP in detecting copy-move and splicing forgeries. Multi-WLD also outperforms state-of-the-art passive forgery detection techniques.

In recent years, exercise games have been criticized for not being able to engage their players into levels of physical activity that are high enough to yield health benefits. A major challenge in the design of exergames, however, is that it is difficult to assess the amount of physical activity an exergame yields due to limitations of existing techniques to assess energy expenditure of exergaming activities. With recent advances in commercial depth sensing technology to accurately track players' motions in 3D, we present a technique called Vizical that uses a non-linear regression approach to accurately predict energy expenditure in real-time. Vizical may allow for creating exergames that can report energy expenditure while playing, and whose intensity can be adjusted in real-time to stimulate larger health benefits.

This paper proposes a method for race recognition from face images using local descriptors. The proposed method uses two types of local descriptors: local binary pattern (LBP) and Weber local descriptors (WLD). First, LBP and WLD histograms are obtained separately from blocks of normalized face image. Kruskal-Wallis feature selection technique is applied to the histograms to select the significant bins for race recognition. Then the selected bins from the two histograms are concatenated block by block to produce the final feature set of the face image. Minimum city block distance is used as a classifier. The experiments are conducted using gray scale FERET images with five race groups. Experimental results show that the proposed method has superior race recognition accuracies for all the five race groups compared to LBP and WLD alone.

Vehicle detection from aerial images is becoming an increasingly important research topic in surveillance, traffic monitoring and military applications. The system described in this paper focuses on vehicle detection in rural environments and its applications to oil and gas pipeline threat detection. Automatic vehicle detection by unmanned aerial vehicles (UAV) will replace current pipeline patrol services that rely on pilot visual inspection of the pipeline from low altitude high risk flights that are often restricted by weather conditions. Our research compares a set of feature extraction methods applied for this specific task and four classification techniques. The best system achieves an average 85% vehicle detection rate and 1800 false alarms per flight hour over a large variety of areas including vegetation, rural roads and buildings, lakes and rivers collected during several day time illuminations and seasonal changes over one year.

In this paper, a novel approach is proposed for perceptual grouping and localization of ill-defined curvilinear structures. Our approach builds upon the tensor voting and the iterative voting frameworks. Its efficacy lies on iterative refinements of curvilinear structures by gradually shifting from an exploratory to an exploitative mode. Such a mode shifting is achieved by reducing the aperture of the tensor voting fields, which is shown to improve curve grouping and inference by enhancing the concentration of the votes over promising, salient structures. The proposed technique is validated on delineating adherens junctions that are imaged through fluorescence microscopy. However, the method is also applicable for screening other organisms based on characteristics of their cell wall structures. Adherens junctions maintain tissue structural integrity and cell-cell interactions. Visually, they exhibit fibrous patterns that may be diffused, heterogeneous in fluorescence intensity, or punctate and frequently perceptual. Besides the application to real data, the proposed method is compared to prior methods on synthetic and annotated real data, showing high precision rates.

In this paper, we present a new fast method for matching stereo images acquired by a stereo sensor embedded in a moving vehicle. The method consists in exploiting the matching results obtained in one stereo pair (frame) for computing the disparity map of the following stereo pair. This can be achieved by finding a temporal relationship, which we named association, between consecutive frames. The disparity range of the current frame is deduced from the disparity map of the preceding frame and the association between the two frames. Dynamic programming technique is considered for matching the image features. The proposed approach is tested on virtual and real stereo image sequences and the results are satisfactory. The method is fast and able to provide about 20 millions disparity maps per second on a HP Pavilion dv6700 2.1GHZ.

In this paper the authors present a technique suitable for segmenting high resolution sidescan sonar images, which is a problem that may arise in military or civilian applications. To this end, they use the "active contours without edges" model that was proposed by Chan and Vese.

Understanding intent is an important aspect of communication among people and is an essential component of the human cognitive system. This capability is particularly relevant for situations that involve collaboration among agents or detection of situations that can pose a threat. In this paper, we propose an approach that allows a robot to detect intentions of others based on experience acquired through its own sensory-motor capabilities, then using this experience while taking the perspective of the agent whose intent should be recognized. Our method uses a novel formulation of Hidden Markov Models designed to model a robot's experience and interaction with the world. The robot's capability to observe and analyze the current scene employs a novel vision-based technique for target detection and tracking, using a non-parametric recursive modeling approach. We validate this architecture with a physically embedded robot, detecting the intent of several people performing various activities.

We present a hierarchical feature fusion model for image classification that is constructed by an evolutionary learning algorithm. The model has the ability to combine local patches whose location, width and height are automatically determined during learning. The representational framework takes the form of a two-level hierarchy which combines feature fusion and decision fusion into a unified model. The structure of the hierarchy itself is constructed automatically during learning to produce optimal local feature combinations. A comparative evaluation of different classifiers is provided on a challenging gender classification image database. It demonstrates the effectiveness of these Feature Fusion Hierarchies (FFH).

Background modeling is an essential and important part of many high-level video processing applications. Recently, the Support Vector Data Description (SVDD) has been introduced for novelty detection when only one class of data is available, i.e. background pixels. This paper proposes a method to efficiently train an SVDD and compares the performance of this training algorithm with the traditional SVDD training techniques. We compare the performance of our method with traditional SVDD and other classification algorithms on various data sets including real video sequences.

We consider the problem of recognizing 3-D objects from 2-D images using geometric models and assuming different viewing angles and positions. Our goal is to recognize and localize instances of specific objects (i.e., model-based) in a scene. This is in contrast to category-based object recognition methods where the goal is to search for instances of objects that belong to a certain visual category (e.g., faces or cars). The key contribution of our work is improving 3-D object recognition by integrating Algebraic Functions of Views (AFoVs), a powerful framework for predicting the geometric appearance of an object due to viewpoint changes, with indexing and learning. During training, we compute the space of views that groups of object features can produce under the assumption of 3-D linear transformations, by combining a small number of reference views that contain the object features using AFoVs. Unrealistic views (e.g., due to the assumption of 3-D linear transformations) are eliminated by imposing a pair of rigidity constraints based on knowledge of the transformation between the reference views of the object. To represent the space of views that an object can produce compactly while allowing efficient hypothesis generation during recognition, we propose combining indexing with learning in two stages. In the first stage, we sample the space of views of an object sparsely and represent information about the samples using indexing. In the second stage, we build probabilistic models of shape appearance by sampling the space of views of the object densely and learning the manifold formed by the samples. Learning employs the Expectation-Maximization (EM) algorithm and takes place in a "universal," lower-dimensional, space computed through Random Projection (RP). During recognition, we extract groups of point features from the scene and we use indexing to retrieve the most feasible model groups that might have produced them (i.e., hypothesis generation). The likelihood of each hypothesis is then computed using the probabilistic models of shape appearance. Only hypotheses ranked high enough are considered for further verification with the most likely hypotheses verified first. The proposed approach has been evaluated using both artificial and real data, illustrating promising performance. We also present preliminary results illustrating extensions of the AFoVs framework to predict the intensity appearance of an object. In this context, we have built a hybrid recognition framework that exploits geometric knowledge to hypothesize the location of an object in the scene and both geometrical and intesnity information to verify the hypotheses.

This paper proposes an automatic framework for figure-ground segmentation of edged images in the presence of cluttered background. Our work employs perceptual grouping concepts to characterize image segments by means of their saliency, which is computed via tensor voting. The main innovation of our work is a case-based thresholding scheme which iteratively eliminates edge segments with low-saliency in multiple scales, preserving those that are more likely to belong to foreground. The key idea is classifying saliency histograms in several cases by considering the relative position of the modes of the figure/ground distributions and applying specific actions in each case. We have performed extensive experiments in order to evaluate our framework both quantitatively and qualitatively, including real images from the Berkeley dataset. 1

This paper describes a novel hand-based verification system based on palm-finger segmentation and fusion. The proposed system operates on 2D hand images acquired by placing the hand on a planar lighting table without any guidance pegs. The segmentation of the palm and the fingers is performed without requiring the extraction of any landmark points on the hand. First, the hand is segmented from the forearm using a robust, iterative methodology based on morphological operators. Then, the hand is segmented into six regions corresponding to the palm and the fingers using morphological operators again. The geometry of each component of the hand is represented using high order Zernike moments which are computed using an efficient methodology. Finally, verification is performed by fusing information from different parts of the hand. The proposed system has been evaluated on a database of 101 subjects illustrating high accuracy and robustness. Comparisons with competitive approaches that use the whole hand illustrate the superiority of the proposed, component-based, approach both in terms of accuracy and robustness. Qualitative comparisons with state of the art systems illustrate that the proposed system has comparable or better performance.

Recognizing human actions from video has been a challenging problem in computer vision. Although human actions can be inferred from a wide range of data, it has been demonstrated that simple human actions can be inferred by tracking the movement of the head in 2D. This is a promising idea as detecting and tracking the head is expected to be simpler and faster because the head has lower shape variability and higher visibility than other body parts (e.g., hands and/or feet). Although tracking the movement of the head alone does not provide sufficient information for distinguishing among complex human actions, it could serve as a complimentary component of a more sophisticated action recognition system. In this article, we extend this idea by developing a more general, viewpoint invariant, action recognition system by detecting and tracking the 3D position of the head using multiple cameras. The proposed approach employs Principal Component Analysis (PCA) to register the 3D trajectories in a common coordinate system and Dynamic Time Warping (DTW) to align them in time for matching. We present experimental results to demonstrate the potential of using 3D head trajectory information to distinguish among simple but common human actions independently of viewpoint.

Representing, associating, interpreting and learning behavioral patterns and rules from observation is always a challenging research problem for cognitive systems and human computer interaction. This problem becomes more difficult in complex environments with many unexpected behavioral patterns from objects, or cells, or individuals or small group of people acting under certain plans at the same time in different locations. This paper targets the development of a novel synergistic methodology for automatically monitoring, representing, associating and interpreting behavioral patterns and activities, and learning rules from complex environments using videos

Developing on-board automotive driver assistance systems aiming to alert drivers about driving environments, and possible collision with other vehicles has attracted a lot of attention lately. In these systems, robust and reliable vehicle detection is a critical step. This paper presents a review of recent vision-based on-road vehicle detection systems. Our focus is on systems where the camera is mounted on the vehicle rather than being fixed such as in traffic/driveway monitoring systems. First, we discuss the problem of on-road vehicle detection using optical sensors followed by a brief review of intelligent vehicle research worldwide. Then, we discuss active and passive sensors to set the stage for vision-based vehicle detection. Methods aiming to quickly hypothesize the location of vehicles in an image as well as to verify the hypothesized locations are reviewed next. Integrating detection with tracking is also reviewed to illustrate the benefits of exploiting temporal continuity for vehicle detection. Finally, we present a critical overview of the methods discussed, we assess their potential for future deployment, and we present directions for future research.

Robust and reliable vehicle detection from images acquired by a moving vehicle (i.e., on-road vehicle detection) is an important problem with applications to driver assistance systems and autonomous, self-guided vehicles. The focus of this work is on the issues of feature extraction and classification for rear-view vehicle detection. Specifically, by treating the problem of vehicle detection as a two-class classification problem, we have investigated several different feature extraction methods such as principal component analysis, wavelets, and Gabor filters. To evaluate the extracted features, we have experimented with two popular classifiers, neural networks and support vector machines (SVMs). Based on our evaluation results, we have developed an on-board real-time monocular vehicle detection system that is capable of acquiring grey-scale images, using Ford's proprietary low-light camera, achieving an average detection rate of 10 Hz. Our vehicle detection algorithm consists of two main steps: a multiscale driven hypothesis generation step and an appearance-based hypothesis verification step. During the hypothesis generation step, image locations where vehicles might be present are extracted. This step uses multiscale techniques not only to speed up detection, but also to improve system robustness. The appearance-based hypothesis verification step verifies the hypotheses using Gabor features and SVMs. The system has been tested in Ford's concept vehicle under different traffic conditions (e.g., structured highway, complex urban streets, and varying weather conditions), illustrating good performance.

Robust and reliable vehicle detection from images acquired by a moving vehicle is an important problem with numerous applications including driver assistance systems and self-guided vehicles. Our focus in this paper is on improving the performance of on-road vehicle detection by employing a set of Gabor filters specifically optimized for the task of vehicle detection. This is essentially a kind of feature selection, a critical issue when designing any pattern classification system. Specifically, we propose a systematic and general evolutionary Gabor filter optimization (EGFO) approach for optimizing the parameters of a set of Gabor filters in the context of vehicle detection. The objective is to build a set of filters that are capable of responding stronger to features present in vehicles than to nonvehicles, therefore improving class discrimination. The EGFO approach unifies filter design with filter selection by integrating genetic algorithms (GAs) with an incremental clustering approach. Filter design is performed using GAs, a global optimization approach that encodes the Gabor filter parameters in a chromosome and uses genetic operators to optimize them. Filter selection is performed by grouping filters having similar characteristics in the parameter space using an incremental clustering approach. This step eliminates redundant filters, yielding a more compact optimized set of filters. The resulting filters have been evaluated using an application-oriented fitness criterion based on support vector machines. We have tested the proposed framework on real data collected in Dearborn, MI, in summer and fall 2001, using Ford's proprietary low-light camera.

We present the design and implementation of an image analysis system for the automatic analysis of Heliothis zea insect images. The Heliothis zea is a corn earworm eating corn crops. Biotech researchers are interested in developing insecticidal bio-toxins with the best performance to kill or stunt the growth of this insects. The automated analysis of Heliothis zea images is imperative for fast and efficient biotech experiments. The proposed system consists of three stages: (i) insect segmentation, (ii) region processing, and (iii) instar and life classification. A probabilistic model (PM) based on mixtures of Gaussians has shown better performance for segmenting the insect images. And a back-propagation neural network (NN) has shown better performance for classifying the insect instar stage. The proposed system has been evaluated on real data using a fivefold cross validation procedure.

Feature subset selection has received considerable attention in the machine learning literature, however, it has not been fully explored or exploited in the computer vision area. In this paper, we consider the problem of object detection using genetic algorithms (GA) for feature subset selection. We argue that feature selection is an important problem in object detection, and demonstrate that GA provide a simple, general, and powerful framework for selecting good sets of features, leading to lower detection error rates. As a case study, we have chosen to perform feature extraction using the popular method of principal component analysis (PCA) and classification using support vector machines (SVM). We have tested this framework on. two difficult and practical object detection problems: vehicle detection and face detection. Experimental results demonstrate significant performance improvements in both cases.

We present a distributed vision-based surveillance system. The system acquires and processes grey level images through one or multiple camera units monitoring certain area(s) via a local area network (LAN) and is capable of combining information from multiple camera units to obtain a consensus decision. It can be trained to detect certain type of intrusions, for example pedestrians, a group of pedestrians, vehicles, pets, etc., and minimizes false alerts due to other non-interested intrusions. As a case study, we aim to detect pedestrian/vehicle in an observation area. Our vision-based intrusion detection approach consists of two main steps: background subtraction based hypothesis generation (HG) and appearance-based hypothesis verification (HV). HG hypothesizes possible threats (intrusions), and HV verifies those hypotheses using a Gabor filter for feature extraction and support vector machines (SVMs) for classification. The system has been tested in an unconstrained outdoor environment, illustrating good performance.

Despite the considerable amount of research work on the application of Gabor filters in pattern classification, their design and selection have been mostly done on a trial and error basis. Existing techniques are either only suitable for a small number of filters or less problem-oriented. A systematic and general evolutionary Gabor filter optimization (EGFO) approach that yields a more optimal, problem-specific, set of filters is proposed in this study. The EGFO approach unifies filter design with filter selection by integrating genetic algorithms (GAs) with an incremental clustering approach. Specifically, filter design is performed using GAs, a global optimization approach that encodes the parameters of the Gabor filters in a chromosome and uses genetic operators to optimize them. Filter selection is performed by grouping together filters having similar characteristics (i.e., similar parameters) using incremental clustering in the parameter space. Each group of filters is represented by a single filter whose parameters correspond to the average parameters of the filters in the group. This step eliminates redundant filters, leading to a compact, optimized set of filters. The average filters are evaluated using an application-oriented fitness criterion based on support vector machines (SVMs). To demonstrate the effectiveness of the proposed framework, we have considered the challenging problem of vehicle detection from gray-scale images. Our experimental results illustrate that the set of Gabor filters, specifically optimized for the problem of vehicle detection, yield better performance than using traditional filter banks.

We discuss and present a computer vision system that monitors a sitting baby of ages of 6 months old to one-year-old. The system tracks a baby's mouth and hands and alarms the guardians if it finds some objects in the baby's hand going towards his or her mouth. We use skin detection, grayscale based sum of square differences (SSD) and template matching for tracking the hands and the head of a baby.

The focus of this work is on the problem of feature selection and classification for on-road vehicle detection. In particular, we propose using quantized Haar wavelet features and Support Vector Machines (SVMs) for rear-view vehicle detection. Wavelet features are particularly attractive for vehicle detection because they form a compact representation, encode edge information, capture information from multiple scales, and can be computed efficiently. Traditionally, methods using wavelet features for classification truncate the coefficients by keeping only the ones with largest magnitude. We believe that the actual values of the wavelet coefficients are not very important for vehicle detection. In fact, the coefficient magnitudes indicate local oriented intensity differences, information that cold be very different even for the same vehicle under different lighting conditions. Therefore, we argue and demonstrate experimentally that the actual coefficient values are less important compared to the simple presence or absence of those coefficients. Specifically, we propose quantizing large negative coefficients to -1, large positive coefficients to 1, and setting the rest coefficients to 0. The quantized coefficients seem to encode important information about the general shape and structure of vehicles, while ignoring fine details and allowing for sufficient inter-class variability. Experimental results and comparisons using real data demonstrate the superiority of the proposed approach which has achieved an average accuracy of 93.94% on completely novel test images.

We consider the problem of gender classification from frontal facial images using genetic feature subset selection. We argue that feature selection is an important issue in gender classification and demonstrate that Genetic Algorithms (GA) can select good subsets of features (i.e., features that encode mostly gender information), reducing the classification error. First, Principal Component Analysis (PCA) is used to represent each image as a feature vector (i.e., eigen-features) in a low-dimensional space. Genetic Algorithms (GAs) are then employed to select a subset of features from the low-dimensional representation by disregarding certain eigenvectors that do not seem to encode important gender information. Four different classifiers were compared in this study using genetic feature subset selection: a Bayes classifier, a Neural Network (NN) classifier, a Support Vector Machine (SVM) classifier, and a classifier based on Linear Discriminant Analysis (LDA). Our experimental results show a significant error rate reduction in all cases. The best performance was obtained using the SVM classifier. Using only 8.4% of the features in the complete set, the SVM classifier achieved an error rate of 4.7% from an average error rate of 8.9% using manually selected features.

This paper presents an in-vehicle real-time monocular precrash vehicle detection system. The system acquires grey level images through a forward facing low light camera and achieves an average detection rate of 10Hz. The vehicle detection algorithm consists of two main steps: multi-scale driven hypothesis generation and appearance-based hypothesis verification. In the multi-scale hypothesis generation step, possible image locations where vehicles might be present are hypothesized. This step uses multi-scale techniques to speed up detection but also to improve system robustness by making system performance less sensitive to the choice of certain parameters. Appearance-base hypothesis verification verifies those hypothesis using Haar Wavelet decomposition for feature extraction and Support Vector Machines (SVMs) for classification. The monocular system was tested under different traffic scenarios (e.g., simply structured highway, complex urban street, varying weather conditions), illustrating good performance.

Investigates the application of genetic algorithms (GAs) for recognizing real 2D or 3D objects from 2D intensity images, assuming that the viewpoint is arbitrary. Our approach is model-based (i.e. we assume a pre-defined set of models), while our recognition strategy relies on the theory of algebraic functions of views. According to this theory, the variety of 2D views depicting an object can be expressed as a combination of a small number of 2D views of the object. This implies a simple and powerful strategy for object recognition: novel 2D views of an object (2D or 3D) can be recognized by simply matching them to combinations of known 2D views of the object. In other words, objects in a scene are recognized by "predicting" their appearance through the combination of known views of the objects. This is an important idea, which is also supported by psychophysical findings indicating that the human visual system works in a similar way. The main difficulty in implementing this idea is determining the parameters of the combination of views. This problem can be solved either in the space of feature matches among the views ("image space") or the space of parameters ("transformation space"). In general, both of these spaces are very large, making the search very time-consuming. In this paper, we propose using GAs to search these spaces efficiently. To improve the efficiency of genetic searching in the transformation space, we use singular value decomposition and interval arithmetic to restrict the genetic search to the most feasible regions of the transformation space. The effectiveness of the GA approaches is shown on a set of increasingly complex real scenes where exact and near-exact matches are found reliably and quickly.

This paper describes a new approach for representing, recognizing and interpreting human activity from video. The approach presented (at the conceptual level) is a model based on the hierarchical synergy of three other models (the L-G graph, the SPN graph and a NN model). In particular, in our project human activity is strongly related with the ability of describing and interrelating events. Thus, the L-G graph (local-global graph) provides a powerful description of the structural image features presented in an event, the SPN (stochastic Petri net) model offers a description of the functional behavior of the changes or operations in video presented in an event, and the NN (neural network) model provides the capability of extracting and learning behavioral patterns, presented in human activities.

In this paper we introduce a procedure, based on the max-min clustering method, that identifies a fixed order of training pattern presentation for fuzzy adaptive resonance theory mapping (ARTMAP). This procedure is referred to as the ordering algorithm, and the combination of this procedure with fuzzy ARTMAP is referred to as ordered fuzzy ARTMAP. Experimental results demonstrate that ordered fuzzy ARTMAP exhibits a generalization performance that is better than the average generalization performance of fuzzy ARTMAP, and in certain cases as good as, or better than the best fuzzy ARTMAP generalization performance. We also calculate the number of operations required by the ordering algorithm and compare it to the number of operations required by the training phase of fuzzy ARTMAP. We show that, under mild assumptions, the number of operations required by the ordering algorithm is a fraction of the number of operations required by fuzzy ARTMAP.

Current indexing-based approaches build the hash table using either a large number of reference views or 3D models. In this paper, we propose building the hash table using algebraic functions of views. During preprocessing, we consider groups of model points and we represent all the views (i.e., images) that they can produce in a hash table. These views are computed using algebraic functions of a small number of reference views which contain the group. Fundamental to this procedure is a methodology based on singular value decomposition and interval arithmetic for estimating the ranges of values that the parameters of algebraic functions can assume. During recognition, scene groups are used to retrieve from the hash table the model groups that might have produced them. Using algebraic functions of views for indexing-based recognition offers a number of advantages. First of all, the hash table can be built easier without requiring 3D models or a large number of reference views. Second, recognition does not rely on the similarities between new and reference views. Third, verification becomes simpler. Finally, the approach is more general and extendible.

A major problem associated with geometric hashing and methods which have emerged from it is the nonuniform distribution of invariants over the hash space. This has two serious effects on the performance of the method. First, it can result in an inefficient storage of data which can increase recognition time. Second, given that geometric hashing is highly amenable to parallel implementation, a nonuniform distribution of data poses difficulties in tackling the load-balancing problem. Finding a "good" geometric hash function which redistributes the invariants uniformly over the hash space is not easy. Current approaches make assumptions about the statistical characteristics of the data and then use techniques from probability theory to calculate a transformation that maps the nonuniform distribution of invariants to a uniform one. In this paper, a new approach is proposed based on an elastic hash table. In contrast to existing approaches which try to redistribute the invariants over the hash bins, we proceed oppositely by distributing the hash bins over the invariants. The key idea is to associate the hash bins with the output nodes of a self-organizing feature map (SOFM) neural network which is trained using the invariants as training examples. In this way, the location of a hash bin in the space of invariants is determined by the weight vector of the node associated with the hash bin. During training, the SOFM spreads the hash bins proportionally to the distribution of invariants (i.e., more hash bins are assigned to higher density areas while less hash bins are assigned to lower density areas) and adjusts their size so that they eventually hold almost the same number of invariants. The advantage of the proposed approach is that it is a process that adapts to the invariants through learning. Hence, it makes absolutely no assumptions about the statistical characteristics of the invariants and the geometric hash function is actually computed through learning. Furthermore, SOFM's "topology preserving" property ensures that the computed geometric hash function should be well behaved. The proposed approach, was shown to perform well on both artificial and real data.

We consider the problem of learning the mapping between the image coordinates of unknown affine views of an object and the parameters of the affine transformation that can align a known view of the same object with them. A single layer neural network (SL-NN) is used to learn the mapping. Although the proposed approach is conceptually similar to other approaches in the literature, its practical advantages are more profound. The views used to train the SL-NN are not obtained by taking different pictures of the object but by sampling the space of its affine transformed views. This space is constructed by estimating the range of values that the parameters of affine transformation can assume using a single view and a methodology based on singular value decomposition and interval arithmetic. The proposed scheme is as accurate as traditional least-squares approaches but faster. A front-end stage to the SL-NN, based on principal components analysis increases its noise tolerance dramatically and guides us in deciding how many training views are necessary in order for it to learn a good mapping.

A major problem associated with geometric hashing and methods which have emerged from it is the non-uniform distribution of invariants over the hash space. This problem can affect the performance of the method significantly. Finding a "good" geometric hash function which redistributes the invariants uniformly over the hash space is not easy. In this paper, a new approach is proposed for alleviating the above problem. It is based on the use of an "elastic hash table" which is implemented as a self-organizing feature map neural network (SOFM-NN). In contrast to existing approaches which try to redistribute the invariants over the hash bins, we proceed oppositely, spreading the hash bins over the invariants. During training, the SOFM-NN resembles an elastic net which deforms over the hash space. The objective of the deformation process is to spread more hash bins in hash space areas which are heavily occupied and less hash bins in lower density areas. The advantage of the proposed approach is that it is a process that adapts to the invariants through learning. Hence, it makes absolutely no assumptions about the statistical characteristics of the invariants and the geometric hash function is actually computed through learning. Furthermore, the well known "topology preserving" property of the SOFM-NN guarantees that the computed geometric hash function should be well behaved. Finally, the proposed approach is inherently parallelizable.>