Showing papers on "Monocular vision published in 2008"

3-D Depth Reconstruction from a Single Still Image

Abstract: We consider the task of 3-d depth estimation from a single still image. We take a supervised learning approach to this problem, in which we begin by collecting a training set of monocular images (of unstructured indoor and outdoor environments which include forests, sidewalks, trees, buildings, etc.) and their corresponding ground-truth depthmaps. Then, we apply supervised learning to predict the value of the depthmap as a function of the image. Depth estimation is a challenging problem, since local features alone are insufficient to estimate depth at a point, and one needs to consider the global context of the image. Our model uses a hierarchical, multiscale Markov Random Field (MRF) that incorporates multiscale local- and global-image features, and models the depths and the relation between depths at different points in the image. We show that, even on unstructured scenes, our algorithm is frequently able to recover fairly accurate depthmaps. We further propose a model that incorporates both monocular cues and stereo (triangulation) cues, to obtain significantly more accurate depth estimates than is possible using either monocular or stereo cues alone.

Attentional Landmarks and Active Gaze Control for Visual SLAM

Abstract: This paper is centered around landmark detection, tracking, and matching for visual simultaneous localization and mapping using a monocular vision system with active gaze control. We present a system that specializes in creating and maintaining a sparse set of landmarks based on a biologically motivated feature-selection strategy. A visual attention system detects salient features that are highly discriminative and ideal candidates for visual landmarks that are easy to redetect. Features are tracked over several frames to determine stable landmarks and to estimate their 3-D position in the environment. Matching of current landmarks to database entries enables loop closing. Active gaze control allows us to overcome some of the limitations of using a monocular vision system with a relatively small field of view. It supports 1) the tracking of landmarks that enable a better pose estimation, 2) the exploration of regions without landmarks to obtain a better distribution of landmarks in the environment, and 3) the active redetection of landmarks to enable loop closing in situations in which a fixed camera fails to close the loop. Several real-world experiments show that accurate pose estimation is obtained with the presented system and that active camera control outperforms the passive approach.

Large-Scale SLAM Building Conditionally Independent Local Maps: Application to Monocular Vision

Abstract: Simultaneous localization and mapping (SLAM) algorithms based on local maps have been demonstrated to be well suited for mapping large environments as they reduce the computational cost and improve the consistency of the final estimation. The main contribution of this paper is a novel submapping technique that does not require independence between maps. The technique is based on the intrinsic structure of the SLAM problem that allows the building of submaps that can share information, remaining conditionally independent. The resulting algorithm obtains local maps in constant time during the exploration of new terrain and recovers the global map in linear time after simple loop closures without introducing any approximations besides the inherent extended Kalman filter linearizations. The memory requirements are also linear with the size of the map. As the algorithm works in a covariance form, well-known data-association techniques can be used in the usual manner. We present experimental results using a handheld monocular camera, building a map along a closed-loop trajectory of 140 m in a public square, with people and other clutter. Our results show that the combination of conditional independence, which enables the system to share the camera and feature states between submaps, and local coordinates, which reduce the effects of linearization errors, allow us to obtain precise maps of large areas with pure monocular SLAM in real time.

Perceptuo-Motor Interactions during Prehension Movements

Abstract: Adaptive behavior relies on the integration of perceptual and motor processes. In this study, we aimed at characterizing the cerebral processes underlying perceptuo-motor interactions evoked during prehension movements in healthy humans, as measured by means of functional magnetic resonance imaging. We manipulated the viewing conditions (binocular or monocular) during planning of a prehension movement, while parametrically varying the slant of the grasped object. This design manipulates the relative relevance and availability of different depth cues necessary for accurate planning of the prehension movement, biasing visual information processing toward either the dorsal visual stream (binocular vision) or the ventral visual stream (monocular vision). Two critical nodes of the dorsomedial visuomotor stream [V6A (anterior visual area 6) and PMd (dorsal premotor cortex)] increased their activity with increasing object slant, regardless of viewing conditions. In contrast, areas in both the dorsolateral visuomotor stream [anterior intraparietal area (AIP) and ventral premotor cortex (PMv)] and in the ventral visual stream [lateral-occipital tactile-visual area (LOtv)] showed differential slant-related responses, with activity increasing when monocular viewing conditions and increasing slant required the processing of pictorial depth cues. These conditions also increased the functional coupling of AIP with both LOtv and PMv. These findings support the view that the dorsomedial stream is automatically involved in processing visuospatial parameters for grasping, regardless of viewing conditions or object characteristics. In contrast, the dorsolateral stream appears to adapt motor behavior to the current conditions by integrating perceptual information processed in the ventral stream into the prehension plan.

Fusing Monocular Information in Multicamera SLAM

Abstract: This paper explores the possibilities of using monocular simultaneous localization and mapping (SLAM) algorithms in systems with more than one camera. The idea is to combine in a single system the advantages of both monocular vision (bearings-only, infinite range observations but no 3-D instantaneous information) and stereovision (3-D information up to a limited range). Such a system should be able to instantaneously map nearby objects while still considering the bearing information provided by the observation of remote ones. We do this by considering each camera as an independent sensor rather than the entire set as a monolithic supersensor. The visual data are treated by monocular methods and fused by the SLAM filter. Several advantages naturally arise as interesting possibilities, such as the desynchronization of the firing of the sensors, the use of several unequal cameras, self-calibration, and cooperative SLAM with several independently moving cameras. We validate the approach with two different applications: a stereovision SLAM system with automatic self-calibration of the rig's main extrinsic parameters and a cooperative SLAM system with two independent free-moving cameras in an outdoor setting.

Binocular vision and motion-in-depth.

Abstract: When an object moves in three dimensions, the two eyes' views of the world deliver slightly different information to the visual system, providing binocular cues to depth and motion-in-depth. This short review describes the two main sources of binocular information, namely, changing disparity over time and interocular velocity differences; this could be used for the perception of motion-in-depth. We discuss the evidence obtained in recent years on the extent to which each of them is used in human vision. We also highlight outstanding questions and issues in the field that have yet to be addressed.

IMMPDA vehicle tracking system using asynchronous sensor fusion of radar and vision

Abstract: This paper focuses on recognition and tracking of maneuvering vehicles in dense traffic situations. We present an asynchronous multi obstacle multi sensor tracking method that fuses information from radar and monocular vision. A low level fusion method is integrated into the framework of an IMMPDA Kalman filter. Real world experiments demonstrate that the system combines the complementary strengths of the employed sensors.

Vestibulo-ocular responses to vertical translation in normal human subjects

Abstract: Prior studies of the human translational vestibulo-ocular reflex (tVOR) report that eye rotations amount to less than 60% of those required to keep the eyes pointed at a stationary visual target, unlike the angular VOR (aVOR) which is optimized to maintain stable gaze. Our first goal was to determine if the performance of the tVOR improves when head translations are combined with head rotations in ambient lighting. A second goal was to measure tVOR during vertical head translations (bob), which has not received systematic study. We measured tVOR alone and in combination with the aVOR in 20 normal human subjects, aged 25–72 years, as they sat on a moving platform that bobbed at 2.0 Hz while rotating horizontally (yaw) at 1.0 Hz. When subjects viewed a visual target at 2 m, median “compensation gain” (eye rotational velocity/required eye rotational velocity to maintain foveal target fixation) was 0.52 during pure bob and 0.59 during combined bob–yaw; during viewing of a near target at ∼17 cm, compensation gain was 0.58 for pure bob and 0.60 for combined bob–yaw. Mean phase lag of eye-in-head velocity for the tVOR was ∼19° with respect to the ideal compensatory response, irrespective of whether translation was accompanied by rotation. Thus, the tVOR changed only slightly during translation–rotation versus pure translation, and our subjects’ ocular rotations remained at about 60% of those required to point the eyes at the target. Comparison of response during binocular or monocular viewing, and ambient or reduced illumination, indicated that relative image motion between the target and background was an important determinant of tVOR behavior. We postulate that tVOR evolved not to stabilize the image of the target on the fovea, but rather to minimize retinal image motion between objects lying in different planes, in order to optimize motion parallax information.

Mono-vision corner SLAM for indoor navigation

Abstract: We present a real-time monocular vision based range measurement method for Simultaneous Localization and Mapping (SLAM) for an Autonomous Micro Aerial Vehicle (MAV) with significantly constrained payload. Our navigation strategy assumes a GPS denied manmade environment, whose indoor architecture is represented via corner based feature points obtained through a monocular camera. We experiment on a case study mission of vision based path-finding through a conventional maze of corridors in a large building.

Monocular vision simultaneous localization and mapping using SURF

Abstract: In visual SLAM features are extracted from images as landmarks. Features should be easily and fast extracted and should be reliably matched in different situations. However, most point features extraction methods used in visual SLAM cannot trade off the reliability of matching and the speed of extraction. In our monocular vision SLAM research, a new feature extractor SURF, which provides both robust matching and high speed extraction, is applied to detect point landmarks from image of indoor environment. We use EKF method to estimate the states of camera and landmarks. In addition, both a new feature detecting strategy which considers landmark distribution in the environment, and an improved landmark management mechanism which allows some landmarks existing permanently to deal with unobservable situation are proposed. The experiments demonstrate that our SLAM approach is effective and has real-time performance.

Hybrid laser and vision based object search and localization

Abstract: We describe a method for an autonomous robot to efficiently locate one or more distinct objects in a realistic environment using monocular vision. We demonstrate how to efficiently subdivide acquired images into interest regions for the robot to zoom in on, using receptive field cooccurrence histograms. Objects are recognized through SIFT feature matching and the positions of the objects are estimated. Assuming a 2D map of the robot's surroundings and a set of navigation nodes between which it is free to move, we show how to compute an efficient sensing plan that allows the robot's camera to cover the environment, while obeying restrictions on the different objects' maximum and minimum viewing distances. The approach has been implemented on a real robotic system and results are presented showing its practicability and the quality of the position estimates obtained.

Methods for Localization and Mapping Using Vision and Inertial Sensors

Abstract: The problems of vision-based localization and mapping are currently highly active areas of research for aerial systems. With a wealth of information available in each image, vision sensors allow vehicles to gather data about their surrounding environment in addition to inferring own-ship information. However, algorithms for processing camera images are often cumbersome for the limited computational power available onboard many unmanned aerial systems. This paper therefore investigates a method for incorporating an inertial measurement unit together with a monocular vision sensor to aid in the extraction of information from camera images, and hence reduce the computational burden for this class of platforms. Feature points are detected in each image using a Harris corner detector, and these feature measurements are statistically corresponded across each captured image using knowledge of the vehicle’s pose. The investigated methods employ an Extended Kalman Filter framework for estimation. Real-time hardware results are presented using a baseline configuration in which a manufactured target is used for generating salient feature points, and vehicle pose information is provided by a high precision motion capture system for comparison purposes.

MVCSLAM: Mono-Vision Corner SLAM for Autonomous Micro-Helicopters in GPS Denied Environments

Abstract: We present a real-time vision navigation and ranging method (VINAR) for the purpose of Simultaneous Localization and Mapping (SLAM) using monocular vision. Our navigation strategy assumes a GPS denied unknown environment, whose indoor architecture is represented via corner based feature points obtained through a monocular camera. We experiment on a case study mission of vision based SLAM through a conventional maze of corridors in a large building with an autonomous Micro Aerial Vehicle (MAV). We propose a method for gathering useful landmarks from a monocular camera for SLAM use. We make use of the corners by exploiting the architectural features of the manmade indoors.

Novel Rao-Blackwellized Particle Filter for Mobile Robot SLAM Using Monocular Vision

Abstract: This paper presents the novel Rao-Blackwellised particle filter (RBPF) for mobile robot simultaneous localization and mapping (SLAM) using monocular vision. The particle filter is combined with unscented Kalman filter (UKF) to extending the path posterior by sampling new poses that integrate the current observation which drastically reduces the uncertainty about the robot pose. The landmark position estimation and update is also implemented through UKF. Furthermore, the number of resampling steps is determined adaptively, which seriously reduces the particle depletion problem, and introducing the evolution strategies (ES) for avoiding particle impoverishment. The 3D natural point landmarks are structured with matching Scale Invariant Feature Transform (SIFT) feature pairs. The matching for multi-dimension SIFT features is implemented with a KD-Tree in the time cost of O(log2). Experiment results on real robot in our indoor environment show the advantages of our methods over previous approaches. Keywords—Mobile robot, simultaneous localization and mapping, Rao-Blackwellised particle filter, evolution strategies, scale invariant feature transform.

Vision-Based Approach Angle and Height Estimation for UAV Landing

Abstract: In order to estimate the approach angle and relative height of Unmanned Aircraft Vehicle (UAV) which lands autonomously, a combinational approach of monocular vision and stereo vision is presented. From monocular sequences, vanishing line is extracted by Hough transform and RANSAC algorithm, and then approach angle of UAV is calculated through vanishing line geometry. From stereo sequences, feature-based matching is adopted to gain depth information by extracting Harris corner. With gained approach angle, height of UAV is obtained by 3-D reconstruction. Kalman filter model is built to obtain accurate height by analyzing motion characteristic of UAV. Experimental results show that the proposed algorithm can effectively estimate the approach angle and height, and converge quickly.

Automatic Navigation for A Mobile Robot with Monocular Vision

Abstract: Monocular vision based navigation method has the merits of simple computation and cheap hardware and is promising to realize real time navigation. A monocular vision based navigation method for a mobile robot moving in unknown environment is presented in the paper. By a special installation of a monocular camera on the top of a mobile robot, the method can realize the obstacle detection, distance measurement and path planning based on one single image, so as to realize automatic navigation of a mobile robot in unknown environment. Experiments on the method were done in real environment and it shows that a mobile robot can move automatically and safely under the guide of its monocular vision system.

Influence of selected variables on monocular, interocular, and binocular visual acuity.

Abstract: PURPOSE: The accurate determination of a patient's monocular and binocular visual acuity (VA) is a key feature of any ophthalmic examination. The work reported here was undertaken to inform the acuity assessment procedure by establishing the influence of test sequence, patient gender, and sensori-motor laterality preference on monocular, interocular, and binocular VA values. METHODS: The test protocol required the determination of independent monocular acuities (testing right and left eyes in a randomized sequence), followed by the binocular value, in each of 100 prepresbyopic subjects conforming to specific selection criteria. A set of Landolt-ring logMAR charts for use at 20 ft (6 m) was produced for this task. In addition, the sighting eye and preferred hand of each subject were established using recognized techniques. RESULTS: The order of monocular vision testing was found to have no statistically significant influence on the VA level recorded. The group mean acuity of the left eye exceeded that of the right by 0.005 logMAR units, a difference of no statistical significance. Compared with female subjects, males revealed a consistently, statistically significant higher group mean acuity of about 0.02 logMAR units (> or =1 logMAR chart symbol). However, the clinical significance of this outcome and the relatively superior left acuity might both be regarded as doubtful. Binocular visual resolution was 13% greater than the mean monocular value. Associations between combinations of preferred eye, writing hand, and better-sighted eye were at levels no greater than chance would predict. CONCLUSION: Informed by these results obtained upon visually-normal subjects, the importance of technique and the need to test a patient's VA to threshold is stressed, not only in the context of appropriate clinical case handling but also with regard to accurate record keeping.

A monocular vision based pedestrian detection system for intelligent vehicles

Abstract: Detecting pedestrians in images is a challenging task, especially for the intelligent vehicle environment where there is a moving camera. In this paper, we develop a monocular vision based pedestrian detection system for intelligent vehicles. We propose a two-stage pedestrian detection approach. A full-body pedestrian detector with Haar-like wavelet features and cascade Adaboost classifier (P. Viola and M. Jones, 2001) is trained to generate some pedestrian candidates on the image. We regard pedestrian as assembly of some parts of the body, and train five part detectors with shapelet features (P. Sabzmeydani and G. Mori, 2007) and Adaboost classifier. Each candidate is detected with these part detectors and is verified using detector ensemble (Shengyang Dai et al., 2007). Finally, after the verification, multiple detections are fused with the mean shift method. Experiments show that our system has high performance in detecting pedestrians in different poses, clothing, illumination, occlusion and background.

Response modes influence the accuracy of monocular and binocular reaching movements.

Abstract: The authors manipulated the availability of monocular and binocular vision during the constituent planning and control stages of a goal-directed reaching task. Furthermore, trials were completed with or without online limb vision to determine whether monocular- or binocular-derived ego-motion cues influence the integration of visual feedback for online limb corrections. Results showed that the manipulation of visual cues during movement planning did not influence planning times or overall kinematics. During movement execution, however, binocular reaches--and particularly those completed with online limb vision--demonstrated heightened endpoint accuracy and stability, a finding directly linked to the adoption of a feedback-based mode of reaching control (i.e., online control). In contrast, reaches performed with online monocular vision produced increased endpoint error and instability and demonstrated reduced evidence of feedback-based corrections (i.e., offline control). Based on these results, the authors propose that the combination of static (i.e., target location) and dynamic (i.e., the moving limb) binocular cues serve to specifically optimize online reaching control. Moreover, results provide new evidence that differences in the kinematic and endpoint parameters of binocular and monocular reaches reflect differences in the extent to which the aforementioned engage in online and offline modes of movement control.

Vehicle motion analysis based on a monocular vision system

Abstract: This paper presents a comprehensive methodology for on-road vehicle motion analysis using a monocular vision system. Vehicle motion analysis plays an essential role in various intelligent vehicle applications, such as cruise control, vehicle platooning, and collision avoidance. In this paper, it's proposed to improve the accuracy of vehicle motion analysis by breaking the task into two complementary steps: incoming vehicle detection and vehicle motion analysis. In the vehicle detection, a new vehicle which enters the observation field will be identified by inspecting its vehicle-related features. Once a vehicle is detected, a fine-level motion analysis mechanism is employed to monitor its position and relative speed based on the temporal consistency exploitation. Specifically, a novel 3-D Pulse-Coupled Neural Network (PCNN) model is employed for optical flow calculation and optimization. The improved optical flow is then interpreted to generate reliable vehicle motion estimation. Overall, the proposed method shows excellent performance in terms of both accuracy and efficiency owing to its effective coarse-to-fine processing scheme and multiple-cue consideration. (6 pages)

An Intelligent Real-Time Monocular Vision-Based AGV System for Accurate Lane Detecting

Abstract: An intelligent vision-based automated guided vehicle (AGV) system is presented in this paper. Embedded system and high performance algorithms are designed to accurately detect the artificial guide line and landmarks. Real-time two dimensional image captured by a CCD camera is processed by DSP (digital signal processor), including filtering, segmenting and labeling connected component. Then the vision system calculates relative distance and slope of guide line. In case of docking, the vision system detects artificial landmarks number placed at the side of path. The simulating results of image processing on Matlab software and experimental results on NHV-1 AGV both demonstrate the algorithms are efficient and robust.

4-D Tensor Voting motion segmentation for obstacle detection in autonomous guided vehicle

Abstract: Creating an obstacle detection system is an important challenge to improve safety for road vehicles. A way to meet the industrial cost requirements is to gather a monocular vision sensor. This paper tackles this problem and defines an highly parallelisable image motion segmentation method for taking into account the current evolution of multi processor computer technology. A complete and modular solution is proposed, based on the Tensor Voting framework extended to the 4D space (x, y, dx, dy), where surfaces describe homogeneous moving areas in the image plan.Watershed segmentation is applied on the result to obtain closed boundaries. Cells are then clustered and labeled with respect to planar parallax rigidity constraints. A visual odometry method, based on texture learning and tracking, is used to estimate residual parallax displacement.

Monocular Vision System that Learns with Approximation Spaces

Abstract: This paper introduces a monocular vision system that learns with approximation spaces to control the pan and tilt operations of a digital camera that is tracking a moving target. This monocular vision system has been designed to facilitate inspection by a line-crawling robot that moves along an electric power transmission line. The principal problem considered in this article is how to use various forms of reinforcement learning to control movements of a digital camera. Prior work on the solution to this problem was done by Chris Gaskett using neural Q-learning starting in 1998 and more recently by Gaskett in 2002. However, recent experiments have revealed that both classical target tracking as well as other forms of reinforcement learning control outperform Q-learning. This article considers various forms of the Actor Critic (AC) method to solve the camera movement control problem. Both the conventional AC method as well as a modified AC method that has a built-in run-and-twiddle (RT) control strategy mechanism, are considered in this article. The RT mechanism introduced by Oliver Selfridge in 1981 is an action control strategy, where an organism continues what it has been doing while things are improving (increasing action reward) and twiddles (changes its action strategy) when past actions yield diminishing rewards. In this work, RT is governed by measurements (by a critic) of the degree of overlap between past behaviour patterns and a behavior pattern template representing a standard are carried out within the framework provided by approximation spaces introduced by Zdzisław Pawlak during the early 1980s. This paper considers how to guide reinforcement learning based on knowledge of acceptable behavior patterns. The contribution of this article is an introduction to actor critic learning methods that benefit from approximation spaces in controlling camera movements during target tracking.

Recurrent attacks of vertigo with monocular oscillopsia.

Abstract: A 44-year-old woman presented with a 1-year history of recurrent attacks of vertigo lasting several seconds, associated only with oscillopsia. The attacks subsided when she closed her right eye. General clinical and neurologic …

Visual control of action in step descent

Abstract: Visual guidance of forwards, sideways, and upwards stepping has been investigated, but there is little knowledge about the visuomotor processes underlying stepping down actions. In this study we investigated the visual control of a single vertical step. We measured which aspects of the stepping down movement scaled with visual information about step height, and how this visual control varied with binocular versus monocular vision. Subjects stepped down a single step of variable and unpredictable height. Several kinematic measures were extracted including a new measure, "kneedrop". This describes a transition in the movement of the lower leg, which occurs at a point proportional to step height. In a within-subjects design, measurements were made with either full vision, monocular vision, or no vision. Subjects scaled kneedrop relative to step height with vision, but this scaling was significantly impaired in monocular and no vision conditions. The study establishes a kinematic marker of visually controlled scaling in single-step locomotion which will allow further study of the visuomotor control processes involved in stepping down.

Automobile cruise control method based on monocular vision and implement system thereof

Abstract: The invention discloses a vehicle cruise control method based on monocular vision and a system for realizing thereof. The methods comprises the following steps: a camera is fixedly arranged near a rearview mirror in front of a driver; a monocular image is obtained from the camera; segmentation and automobile detection is carried out for the image; if no car is in the front, a cruise speed is kept or restored and treatment of current frame image data is finished; if a car is detected in the front, distance and speed of the car are calculated; if the distance of the front car is less than the safe distance, appropriate braking is carried out through an automobile braking control system; or the treatment of the current frame image data is finished to switch to the treatment of next frame image data. The method is realized by a system which consists of the camera, an embedded calculation module, a displayer and a loudspeaker. The vehicle cruise control method has the advantages of high security, small calculation complexity, low requirement for hardware configuration, easy butch production and low cost.