Reliability analysis of transform-based stereo matching techniques, and a new matching constraint

Stereo vision is a method of depth perception, in which depth information is inferred from two (or more) images of a scene, taken from different perspectives. Practical applications for stereo vision include aerial photogrammetry, autonomous vehicle guidance, robotics and industrial automation. The initial motivation behind this work was to produce a stereo vision sensor for mining automation applications. For such applications, the input stereo images would consist of close range scenes of rocks. A fundamental problem faced by matching algorithms is the matching or correspondence problem. This problem involves locating corresponding points or features in two images. For this application, speed, reliability, and the ability to produce a dense depth map are of foremost importance. This work implemented a number of areabased matching algorithms to assess their suitability for this application. Area-based techniques were investigated because of their potential to yield dense depth maps, their amenability to fast hardware implementation, and their suitability to textured scenes such as rocks. In addition, two non-parametric transforms, the rank and census, were also compared. Both the rank and the census transforms were found to result in improved reliability of matching in the presence of radiometric distortion - significant since radiometric distortion is a problem which commonly arises in practice. In addition, they have low computational complexity, making them amenable to fast hardware implementation. Therefore, it was decided that matching algorithms using these transforms would be the subject of the remainder of the thesis. An analytic expression for the process of matching using the rank transform was derived from first principles. This work resulted in a number of important contributions. Firstly, the derivation process resulted in one constraint which must be satisfied for a correct match. This was termed the rank constraint. The theoretical derivation of this constraint is in contrast to the existing matching constraints which have little theoretical basis. Experimental work with actual and contrived stereo pairs has shown that the new constraint is capable of resolving ambiguous matches, thereby improving match reliability. Secondly, a novel matching algorithm incorporating the rank constraint has been proposed. This algorithm was tested using a number of stereo pairs. In all cases, the modified algorithm consistently resulted in an increased proportion of correct matches. Finally, the rank constraint was used to devise a new method for identifying regions of an image where the rank transform, and hence matching, are more susceptible to noise. The rank constraint was also incorporated into a new hybrid matching algorithm, where it was combined a number of other ideas. These included the use of an image pyramid for match prediction, and a method of edge localisation to improve match accuracy in the vicinity of edges. Experimental results obtained from the new algorithm showed that the algorithm is able to remove a large proportion of invalid matches, and improve match accuracy.

A Taxonomy of image matching techniques for stereo vision

Stereo vision is a method of depth perception, in which depth information is inferred from two (or more) images of a scene, taken from different perspectives. Applications of stereo vision include aerial photogrammetry, autonomous vehicle guidance, robotics, industrial automation and stereomicroscopy. A key issue in stereo vision is that of image matching, or identifying corresponding points in a stereo pair. The difference in the positions of corresponding points in image coordinates is termed the parallax or disparity. When the orientation of the two cameras is known, corresponding points may be projected back to find the location of the original object point in world coordinates. Matching techniques are typically categorised according to the nature of the matching primitives they use and the matching strategy they employ. This report provides a detailed taxonomy of image matching techniques, including area based, transform based, feature based, phase based, hybrid, relaxation based, dynamic programming and object space methods. A number of area based matching metrics as well as the rank and census transforms were implemented, in order to investigate their suitability for a real-time stereo sensor for mining automation applications. The requirements of this sensor were speed, robustness, and the ability to produce a dense depth map. The Sum of Absolute Differences matching metric was the least computationally expensive; however, this metric was the most sensitive to radiometric distortion. Metrics such as the Zero Mean Sum of Absolute Differences and Normalised Cross Correlation were the most robust to this type of distortion but introduced additional computational complexity. The rank and census transforms were found to be robust to radiometric distortion, in addition to having low computational complexity. They are therefore prime candidates for a matching algorithm for a stereo sensor for real-time mining applications. A number of issues came to light during this investigation which may merit further work. These include devising a means to evaluate and compare disparity results of different matching algorithms, and finding a method of assigning a level of confidence to a match. Another issue of interest is the possibility of statistically combining the results of different matching algorithms, in order to improve robustness.

Gait energy volumes and frontal gait recognition using depth images

Gait energy images (GEIs) and its variants form the basis of many recent appearance-based gait recognition systems. The GEI combines good recognition performance with a simple implementation, though it suffers problems inherent to appearance-based approaches, such as being highly view dependent. In this paper, we extend the concept of the GEI to 3D, to create what we call the gait energy volume, or GEV. A basic GEV implementation is tested on the CMU MoBo database, showing improvements over both the GEI baseline and a fused multi-view GEI approach. We also demonstrate the efficacy of this approach on partial volume reconstructions created from frontal depth images, which can be more practically acquired, for example, in biometric portals implemented with stereo cameras, or other depth acquisition systems. Experiments on frontal depth images are evaluated on an in-house developed database captured using the Microsoft Kinect, and demonstrate the validity of the proposed approach.

A semi-local method for iterative depth-map refinement

We present an iterative hierarchical algorithm for multi-view stereo. The algorithm attempts to utilise as much contextual information as is available to compute highly accurate and robust depth maps. There are three novel aspects to the approach: 1) firstly we incrementally improve the depth fidelity as the algorithm progresses through the image pyramid; 2) secondly we show how to incorporate visual hull information (when available) to constrain depth searches; and 3) we show how to simultaneously enforce the consistency of the depth-map by continual comparison with neighbouring depth-maps. We show that this approach produces highly accurate depth-maps and, since it is essentially a local method, is both extremely fast and simple to implement.

Fast and robust stereo matching algorithms for mining automation

The mining environment, being complex, irregular, and time-varying, presents a challenging prospect for stereo vision. For this application, speed, reliability, and the ability to produce a dense depth map are of foremost importance. This paper evaluates a number of matching techniques for possible use in a stereo vision sensor for mining automation applications. Area-based techniques have been investigated because they have the potential to yield dense maps, are amenable to fast hardware implementation, and are suited to textured scenes. In addition, two nonparametric transforms, namely, rank and census, have been investigated. Matching algorithms using these transforms were found to have a number of clear advantages, including reliability in the presence of radiometric distortion, low computational complexity, and amenability to hardware implementation.

Fast and robust stereo matching algorithms for mining automation

The mining environment, being complex, irregular and time varying, presents a challenging prospect for stereo vision. The objective is to produce a stereo vision sensor suited to close-range scenes consisting primarily of rocks. This sensor should be able to produce a dense depth map within real-time constraints. Speed and robustness are of foremost importance for this investigation. A number of area based matching metrics have been implemented, including the SAD, SSD, NCC, and their zero-meaned versions. The NCC and the zero meaned SAD and SSD were found to produce the disparity maps with the highest proportion of valid matches. The plain SAD and SSD were the least computationally expensive, due to all their operations taking place in integer arithmetic, however, they were extremely sensitive to radiometric distortion. Non-parametric techniques for matching, in particular, the rank and the census transform, have also been investigated. The rank and census transforms were found to be robust with respect to radiometric distortion, as well as being able to produce disparity maps with a high proportion of valid matches. An additional advantage of both the rank and the census transform is their amenability to fast hardware implementation.

Non-parametric techniques for fast and robust stereo matching

The mining environment presents a challenging prospect for stereo vision. Our objective is to produce a stereo vision sensor suited to close-range scenes consisting mostly of rocks. This sensor should produce a dense depth map within real-time constraints. Speed and robustness are of foremost importance for this application. This paper compares a number of stereo matching algorithms in terms of robustness and suitability to fast implementation. These include traditional area-based algorithms, and algorithms based on non-parametric transforms, notably the rank and census transforms. Our experimental results show that the rank and census transforms are robust with respect to radiometric distortion and introduce less computational complexity than conventional area-based matching techniques.

Suitability of non-parametric stereo matching algorithms for mining automation

The mining environment, being complex, irregular and time varying, presents a challenging prospect for stereo vision. For this application, speed, reliability, and the ability to produce a dense depth map are of foremost importance. This paper assesses the suitability of a number of matching techniques for use in a stereo vision sensor for close range scenes consisting primarily of rocks. These include traditional area-based matching metrics, and non-parametric transforms, in particular, the rank and census transforms. Experimental results show that the rank and census transforms exhibit a number of clear advantages over area-based matching metrics, including their low computational complexity, and robustness to certain types of distortion.

Fast and robust stereo matching algorithms for mining automation

The mining environment, being complex, irregular and time varying, presents a challenging prospect for stereo vision. For this application, speed, reliability, and the ability to produce a dense depth map are of foremost importance. This paper evaluates a number of matching techniques for possible use in a stereo vision sensor for mining automation applications. Area-based techniques have been investigated because they have the potential to yield dense maps, are amenable to fast hardware implementation, and are suited to textured scenes. In addition, two non-parametric transforms, namely, the rank and census, have been investigated. Matching algorithms using these transforms were found to have a number of clear advantages, including reliability in the presence of radiometric distortion, low computational complexity, and amenability to hardware implementation.

High altitude stereo visual odometry

Stereo visual odometry has received little investigation in high altitude applications due to the generally poor performance of rigid stereo rigs at extremely small baseline-to-depth ratios. Without additional sensing, metric scale is considered lost and odometry is seen as effective only for monocular perspectives. This paper presents a novel modification to stereo based visual odometry that allows accurate, metric pose estimation from high altitudes, even in the presence of poor calibration and without additional sensor inputs. By relaxing the (typically fixed) stereo transform during bundle adjustment and reducing the dependence on the fixed geometry for triangulation, metrically scaled visual odometry can be obtained in situations where high altitude and structural deformation from vibration would cause traditional algorithms to fail. This is achieved through the use of a novel constrained bundle adjustment routine and accurately scaled pose initializer. We present visual odometry results demonstrating the technique on a short-baseline stereo pair inside a fixed-wing UAV flying at significant height (~30-100m).

Robust scale initialization for long-range stereo visual odometry

Achieving a robust, accurately scaled pose estimate in long-range stereo presents significant challenges. For large scene depths, triangulation from a single stereo pair is inadequate and noisy. Additionally, vibration and flexible rigs in airborne applications mean accurate calibrations are often compromised. This paper presents a technique for accurately initializing a long-range stereo VO algorithm at large scene depth, with accurate scale, without explicitly computing structure from rigidly fixed camera pairs. By performing a monocular pose estimate over a window of frames from a single camera, followed by adding the secondary camera frames in a modified bundle adjustment, an accurate, metrically scaled pose estimate can be found. To achieve this the scale of the stereo pair is included in the optimization as an additional parameter. Results are presented both on simulated and field gathered data from a fixed-wing UAV flying at significant altitude, where the epipolar geometry is inaccurate due to structural deformation and triangulation from a single pair is insufficient. Comparisons are made with more conventional VO techniques where the scale is not explicitly optimized, and demonstrated over repeated trials to indicate robustness.

Tracking people in 3D using position, size and shape

This paper presents a prototype tracking system for tracking people in enclosed indoor environments where there is a high rate of occlusions. The system uses a stereo camera for acquisition, and is capable of disambiguating occlusions using a combination of depth map analysis, a two step ellipse fitting people detection process, the use of motion models and Kalman filters and a novel fit metric, based on computationally simple object statistics. Testing shows that our fit metric outperforms commonly used position based metrics and histogram based metrics, resulting in more accurate tracking of people.