Numerous researchers and organizations continue to refine technologies that will soon allow vehicles to autonomously drive safely from Point A to Point B. Accuracy determining the position of the self-driving vehicle relative to lane markers and road boundaries specifically during inclement weather conditions continues is of primary importance. This paper presents an investigation and associated results where road land boundary markers are detected in conjunction with the ability decipher the horizon when the front view of the vehicle's path is degraded. Degradation of driving scenes can be attributed to such weather conditions as heavy rain, fog, snow, or dust storms. The detection of lane markers and road boundaries is especially important for roads that exhibit severe curves, aggressive uphill slopes and downhill valleys, respectively. Additionally, we present a model to predict deviations from reference distances associated with roads with such design constraints along with some promising experimental results.

This paper aims to develop a vision-based driver assistance system for scene awareness using video frames obtained from a dashboard camera. A saliency image map is devised with features pertinent to the driving scene. This saliency map mimics the human contour and motion sensitive visual perception by extracting spatial, spectral, and temporal information from the input frames and applying entropy driven image-context-feature data fusion. The resultant fusion output comprises high-level descriptors for still segment boundaries and non-stationary object appearance. Following the segmentation and foreground object detection stage, an adaptive maximum likelihood classifier selects road surface regions. The proposed scene driven vision system improves the driver's situational awareness by enabling adaptive road surface classification. As experimental results demonstrate, context-aware low-level to high-level information fusion based on human vision model produces superior segmentation, tracking, and classification results that lead to high- level abstraction of driving scene.

Rural roads present potential risks for drivers. One of them is horizontal curve, which poses higher risk than freeway. This is the major theme for the presented work here aiming to develop a model that predicts risk of curved roads. Major road geometry components associated with curve structure are road slope type being uphill or downhill, road curvature, and curve direction along with vehicle speed as being a critical factor. In this study, cameras mounted in rear and front ends of a vehicle that capture road images are utilized to detect the components emerging risk. This two-view approach is exploited to obtain vehicle speed and bend slope type, whereas curve direction and road curvature are determined by single-view front camera. The proposed approach is leveraged by geometrical derivations using salient visual clues such as vanishing points and road boundary. Additionally, velocity is estimated by reverse-view technique, that is, plane of front view at instance t and the plane of rear view in t+1. Subsequently, overall potential hazard is predicted by assigning weights for each risk components via developed risk estimation model. The proposed model would be an integral part of an advanced driver assistant system by alerting driver about the prominent risk of horizontal curve ahead of time.

In this paper, we propose a new model for 3-car interactive risky behavior of vehicles travelling in front and behind of a driver (overtaken) car. Following distance of vehicles moving in front and at rear end of the car in question plays an important role for overtaking scenario. Moreover, the distance between the car in front and the vehicle following it should be sufficiently long for preventing collision if overtaking is inevitable for the motorist behind the middle subject vehicle. Here, we consider the roles of the vehicles involved in such a scenario. We observe the behaviors of moving vehicles in front and the rear end of the subject car. To this end, front and rear car images are acquired by two cameras and subjected to vertical and horizontal optical flow edge map creation. In classification stage of the optical flow edge map clusters, a motion vector histogram thresholding method is utilized in conjunction with a decision assessment strategy based on the joint Bayesian belief network statistical model. In turn, not only the trajectories of the cars are captured but also joint behavior of three cars over-taken scenario is estimated using the proposed interactive risk model.

Various approaches have been developed for quantifying and displaying network traffic information for determining network status and in detecting anomalies. Although many of these methods are effective, they rely on the collection of long-term network statistics. Here, we present an approach that uses short-term observations of network features and their respective time averaged entropies. Acute changes are localized in network feature space using adaptive Wiener filtering and auto-regressive moving average modeling. The color-enhanced datagram is designed to allow a network engineer to quickly capture and visually comprehend at a glance the statistical characteristics of a network anomaly. First, average entropy for each feature is calculated for every second of observation. Then, the resultant short-term measurement is subjected to first- and second-order time averaging statistics. These measurements are the basis of a novel approach to anomaly estimation based on the well-known Fisher linear discriminant (FLD). Average port, high port, server ports, and peered ports are some of the network features used for stochastic clustering and filtering. We empirically determine that these network features obey Gaussian-like distributions. The proposed algorithm is tested on real-time network traffic data from Ohio University's main Internet connection. Experimentation has shown that the presented FLD-based scheme is accurate in identifying anomalies in network feature space, in localizing anomalies in network traffic flow, and in helping network engineers to prevent potential hazards. Furthermore, its performance is highly effective in providing a colorized visualization chart to network analysts in the presence of bursty network traffic.

In this work, we develop a fast binary partition tree based variable size video coding system. New adaptive algorithms proposed herein are applied to a video encoder with binary partition trees. First, to reduce the computation for block-matching, an adaptive search area method is described which adjusts the searching region according to the size of each block. Second, an early termination method is introduced which terminates the binary partitioning process adaptively according to the statistics of the peak-signal-to-noise-ratio values during each step of block splitting. Third, we put forward a new model for fast rate-distortion (RD) estimation to decrease the computation of matching pursuit (MP) transform coding. Simulation results show that the proposed techniques provide significant gain in computation speed with little or no sacrifice of RD performance, when compared with non-adaptive binary partitioning scheme.

Fast and efficient detection of anomalies is essential for maintaining a robust and secure network. This research presents a method of anomaly detection based on adaptive Wiener filtering of noise followed by ARMA modeling of network flow data. We dynamically calculate noise and traffic signal statistics using network-monitoring metrics for traffic features such as average port, high port, server ports, and peered ports. The underlying approach is tested on near-real-time Internet traffic in the wide-area network (WAN) of Ohio University. The average port feature is determined to be the most informative measure in the estimation process. High port, server ports, and peered ports are used for confirmation of the anomaly detection result. We empirically determine that most of the network features obey Gaussian-like distributions. Experiments reveal that the method is highly effective in predicting anomalies in network traffic flow and preventing any hazard that they may cause.

This paper describes a non-linear IR (infra-red) scene prediction method for range video surveillance and navigation. A Gabor-filter bank is selected as a primary detector for any changes in a given IR range image sequence. The detected ROI (region of interest) involving arbitrary motion is fed to a non-linear Kalman filter for predicting the next scene in time-varying 3D IR video. Potential applications of this research are mainly in indoor/outdoor heat-change based range measurement, synthetic IR scene generation, rescue missions, and autonomous navigation. Experimental results reported herein show that non-linear Kalman filtering-based scene prediction can perform more accurately than linear estimation of future frames in range and intensity driven sensing. The low least mean square error (LMSE), on the average of about 2% using a bank of 8 Gabor filters, also proves the reliability of the IR scene estimator (or predictor) developed in this work.

This paper presents a Kalman filtering approach to predicting 3-D video infrared (IR) scenes as a CMOS multi-coordinate axis sensory-camera mounted on a mobile vehicle moves forward in a controlled environment. Potential applications of this research can be found in indoor/outdoor heat-change based range measurement, synthetic IR scene generation, rescue missions, and autonomous navigation. Experimental results reported herein dictate that linear Kalman filtering based scene prediction accurately estimates future frames in range and intensity sensing. The low least mean square error (LMSE), on the average of 1%, proves the reliability of the approach to IR scene prediction. Currently, the proposed method is devised for piece-wise linear motion of the sensory system as it navigates in hallway or corridor.

In this paper, we present a new digital video watermarking technique using singular value decomposition (SVD) and two-dimensional principal component analysis (2DPCA). The SVD is used to utilized in the spatial domain while the 2DPCA is employed for embedding in the time domain. We demonstrate that the framework based on the combined use of SVD and 2DPCA well-suited to multichannel video with an arbitrary number of signal channels. Our preliminary research results indicate that the proposed method yields watermarked video with little or no perceptible distortion.

Detecting and recognizing objects in unstructured environments is one of the most challenging tasks in computer vision research. We propose an innovative algorithm, called deformable illumination, to address the problem of illumination variance in natural environments. Parallel to the role of deformable shape in object recognition, deformable illumination is designed as an object detection technique. A unified framework presented here integrates both deformable illumination and deformable shape as a simultaneous scheme for object detection and recognition in unstructured environments. Experimental results show the effectiveness of deformable illumination in addressing illumination variance.

As one of the key techniques for futuristic man-machine interface, facial expression analysis has received much attention in recent years. This paper proposes a hierarchical approach to facial expression recognition in image sequences by exploiting both spatial and temporal characteristics within the framework of hierarchical hidden Markov models (HHMMs). Human faces are automatically detected in the maximum likelihood sense. Gabor-wavelet based features are extracted from image sequences to capture the subtle changes of facial expressions. Four prototype emotions; i.e. happiness, anger, fear and sadness, are investigated using the Cohn-Kanade database and an average of 90.98% person-independent recognition rate is achieved. We also demonstrate that HHMMs outperform HMMs for modeling image sequences with multilevel statistical structure.

Facial event mining is one of the key techniques for automatic human face analysis. It plays an important role in human computer interaction. This paper proposes a new approach to facial event recognition by combining active shape models (ASMs) and coupled hidden Markov models (CHMMs). Based on the assumption that a complex facial event can be decomposed into multiple coupled processes, ASMs are used to track global facial features and to decouple pattern attributes for upper and lower faces separately. These two interacting processes are modeled as a CHMM for training and recognition. Four basic facial events are investigated. Preliminary experiments yield consistent results that show the significant advantage of CHMMs over conventional HMMs for facial event mining in video.

The paper proposes an adaptive framework for shape-based image retrieval with relevance feedback. The motivation is to find an adjustable shape representation scheme that can account for feedback information. Relevance feedback is modeled as a dynamic eigenspace decomposition, and is used to classify the database into relevant and irrelevant groups with respect to the query. Eigenvectors of the subspace are updated by optimizing a linear transform with respect to the J3class separability criterion. Experimental results show that the proposed approach can effectively capture a user's perceptual subjectivity

Although significant research effort has been devoted to the natural scene analysis, there hasn't been much work on natural scene synthesis. The research of such nature can help to understand the hidden structure of natural texture and the illumination patterns for classification and segmentation purposes. The common texture synthesis methods are not suitable for natural scene synthesis due to inhomogeneity of the natural textures. In this paper, we propose a new natural scene synthesis framework based on multiple eigenspace decomposition, which is a natural extension of principle component analysis (PCA). Two levels of multiple eigenspaces are used; label level multiple eigenspaces for generating depth map and color level multiple eigenspaces for generating nature textures to fill the label maps. Experimental results are presented to show the effectiveness of the proposed framework.

In many multimedia applications, image processing tools require very high speed implementation of the two-dimensional (2D) discrete cosine transform (DCT). Two 2D-DCT architectures were realized based on the decomposition of 2D-DCT into two one-dimensional (1D) DCT. Each of the resultant 1D-DCT was implemented using a flow-graph (FG) representation of the real part of the Fourier transform for the first architecture and the distributed arithmetic (DA) for the second architecture, respectively. Error simulation was performed for different data word lengths, and average and maximum absolute errors were calculated for both architectures for several possible worst cases. Based on the simulation results, the correct word sizes were determined for both implementations. By describing the FG architecture in VHDL and implementing it in the Xilinx XC4000 series FPGA chip, our results showed that the proposed technique utilized less combinational logic blocks (CLB) as compared to the FG realization in Kumar et al., (1999), and at the same time provided performance increase of up to 28% (190 Mpix/sec).

A recursive cluster detection technique is described for segmenting the color pictures. The proposed method operates in the (L*,a*,b*)-uniform color space and detects image cluster by fitting them some circular-cylindrical decision volumes. This estimates the clusters' distributions in the uniform color space without imposing any constraints on their forms. Boundaries of the decision elements consists of two planes of constant lightness, two cylinders of constant chroma, and two planes of constant hue. They are determined using 1-D histograms of the L*,H/sup 0/,C* cylindrical coordinates of the space. The detected clusters are then correctly isolated from their neighbors by projecting their estimated color distributions onto the Fisher linear discriminant function for 1-D thresholding.>

This paper describes a systematic approach to the design of the segmentation or lower-level processing unit of a general-purpose computer vision system. It is developed within the context of the human visual system and mathematical pattern recognition theory. The eventual goal of the research is to integrate these two concepts to obtain the visually distinct segments of an image, which have vital importance for the interpretation or higher-level processing unit of a vision system. A new computational (pattern recognition) technique is proposed in accordance with the human color perception. It operates in unsupervised mode without using any training prototypes or requiring a priori knowledge about the input scenes. However, it can also be implemented in supervised mode by means of including external knowledge. In order to obtain a set of features most useful for a particular image, a new feature extraction method is devised in the form of a reference feature library. It is a statistical-structural technique which combines the spatial and spectral information in the local image areas.