Vision impairment significantly impacts independence and quality of life. This paper presents iSense, an advanced wearable assistive device for visually impaired users, which integrates RealSense RGB-D cameras, haptic-auditory feedback, and a high-performance processing platform. Key features include obstacle detection, spatial navigation, object recognition, and task assistance in daily environments. Experimental results highlight the potential of iSense to enhance mobility and independence, exemplifying the role of assistive robotics in promoting inclusivity

Prototype-based learning (PbL) using a winner-take-all (WTA) network based on minimum Euclidean distance (ED-WTA) is an intuitive approach to multiclass classification. By constructing meaningful class centers, PbL provides higher interpretability and generalization than hyperplane-based learning (HbL) methods based on maximum inner product (IP-WTA) and can efficiently detect and reject samples that do not belong to any classes. In this article, we first prove the equivalence of IP-WTA and ED-WTA from a representational power perspective. Then, we show that naively using this equivalence leads to unintuitive ED-WTA networks in which the centers have high distances to data that they represent. We propose ±ED-WTA that models each neuron with two prototypes: one positive prototype, representing samples modeled by that neuron, and a negative prototype, representing the samples erroneously won by that neuron during training. We propose a novel training algorithm for the ±ED-WTA network, which cleverly switches between updating the positive and negative prototypes and is essential to the emergence of interpretable prototypes. Unexpectedly, we observed that the negative prototype of each neuron is indistinguishably similar to the positive one. The rationale behind this observation is that the training data that are mistaken for a prototype are indeed similar to it. The main finding of this article is this interpretation of the functionality of neurons as computing the difference between the distances to a positive and a negative prototype, which is in agreement with the BCM theory. Our experiments show that the proposed ±ED-WTA method constructs highly interpretable prototypes that can be successfully used for explaining the functionality of deep neural networks (DNNs), and detecting outlier and adversarial examples.

Geometrical wavelets have already proved their strength in approximation, compression, and denoising of piecewise constant and piecewise linear images. In this paper, we extend this family by introducing planelets toward an effective representation of indoor depth images. It uses a linear fractional model to capture non-linearity of depth values in the planar regions of the output images of Kinect-like sensors. A block-based compression framework based on planelet approximation is then presented, which uses quadtree decomposition along with spatial predictions as an effective intra-coding scheme. Compared with both classical geometric wavelets and some state-of-the-art image coding algorithms, our method provides desirable quality by explicitly representing edges and planar patches.

This paper proposes an object classification framework based on a geometric grammar aimed for mobile robotic applications. The paper first discusses the geometric grammar as a compact representation form for object categories with primitive parts as its constituent elements. The paper then discusses the object classification implemented as parsing of primitive parts. In particular, two approaches are discussed that constrain the search space in order to render the parsing of the primitive parts practical. The two approaches are experimentally verified, first, for a generic object category of chair applied to real range images acquired with a range camera mounted on a mobile robot and, second, for multiple generic object categories applied to synthetic range images. The experimental results show the practicability of the framework.

This paper presents an incremental object part detection algorithm using a particle filter. The method infers object parts from 3D data acquired with a range camera. The range information is quantized and enhanced by local structure to partially cope with considerable measurement noise and distortion. The augmented voxel representation allows the adaptation of known track-before-detect algorithms to infer multiple object parts in a range image sequence even when each single observation does not contain enough information to do the detection. The appropriateness of the method is successfully demonstrated by two experiments for chair legs.

Toward obtaining a compact and multiresolution representation of 2D range scans, a wavelet framework is proposed for encoding an orientation measure called running angle (RA). A new shrinkage algorithm is developed using discrete wavelet transform of the RA signal, which leads to a simplified polyline approximation of the initial scanned points. This approach is evaluated in terms of segmentation of 2D range scans as a line extractor. As a proof of concept, an experiment is performed in our laboratory hallway by a mobile robot equipped with two SICK laser range finders which shows that it is possible to successfully segment raw measurements of the scanner using the proposed approach and obtain proper linear abstraction. Besides a simple, fast heuristic line extraction algorithm is also proposed for the sake of comparison. It is based on thresholding the changes of the incident angles between the laser beam and the vertices of the initial polygon observed by the scanner. Despite its simplicity, this approach performs rather well and can be used in structured environments with low measurement noise. Both approaches are experimentally evaluated and compared with some well known and commonly used line extraction algorithms.

Simple, fast and lightweight SLAM algorithms are necessary in many embedded robotic systems which soon will be used in houses and offices in order to do various service tasks. In this paper the Orthogonal SLAM algorithm is presented as an answer to this need. In continuation of our previous work, the algorithm is extended to generate 3D maps and empirically validated by mapping the long corridor of our lab with the accuracy comparable with hand measured ground truth. The main contribution resides in the idea of reducing the complexity by using orthogonality constraint in indoor environments. This is done by mapping only planes that are parallel or perpendicular to each other which represent the main structure of most indoor environments. Having this assumption, we use an inclined sensor setup (fixed 2D SICK laser range finders) to generate 3D orthogonal maps. The algorithm is extremely fast since in each step it just processes one line of laser measurements.

In this paper, we describe a new approach for the extrinsic calibration of a camera with a 3D laser range finder, that can be done on the fly. This approach does not require any calibration object. Only few point correspondences are used, which are manually selected by the user from a scene viewed by the two sensors. The proposed method relies on a novel technique to visualize the range information obtained from a 3D laser scanner. This technique converts the visually ambiguous 3D range information into a 2D map where natural features of a scene are highlighted. We show that by enhancing the features the user can easily find the corresponding points of the camera image points. Therefore, visually identifying laser- camera correspondences becomes as easy as image pairing. Once point correspondences are given, extrinsic calibration is done using the well-known PnP algorithm followed by a noninear refinement process. We show the performance of our approach through experimental results. In these experiments, we will use an omnidirectional camera. The implication of this method is important because it brings 3D computer vision systems out of the laboratory and into practical use.

Today, lightweight SLAM algorithms are needed in many embedded robotic systems. In this paper the orthogonal SLAM (OrthoSLAM ) algorithm is presented and empirically validated. The algorithm has constant time complexity in the state estimation and is capable to run real-time. The main contribution resides in the idea of reducing the complexity by means of an assumption on the environment. This is done by mapping only lines that are parallel or perpendicular to each other which represent the main structure of most indoor environments. The combination of this assumption with a Kalman filter and a relative map approach is able to map our laboratory hallway with the size of 80 m times 50 m and a trajectory of more than 500 m. The precision of the resulting map is similar to the measurements done by hand which are used as the ground-truth

This paper is concerned with the derivation of the kinematics model of the University of Tehran-Pole Climbing Robot (UT-PCR). As the first step, an appropriate set of coordinates is selected and used to describe the state of the robot. Nonholonomic constraints imposed by the wheels are then expressed as a set of differential equations. By describing these equations in terms of the state of the robot an underactuated driftless nonlinear control system with affine inputs that governs the motion of the robot is derived. A set of experimental results are also given to show the capability of the UT-PCR in climbing a stepped pole.