A cost-benefit analysis of a third camera for stereo correspondence

Abstract

This paper looks at the twin issues of the gain in accuracy of stereo correspondence and the accompanying increase in computational cost due to the use of a third camera for stereo analysis. Trinocular stereo algorithms differ from binocular algorithms essentially in the epipolar constraint used in the local matching stage. The current literature does not provide any insight into the relative merits of binocular and trinocular stereo matching with the matching accuracy being verified against the ground truth. Experiments for evaluating the relative performance of binocular and trinocular stereo algorithms were conducted. The stereo images used for the performance evaluation were generated by applying a Lambertian reflectance model to real Digital Elevation Maps (DEMs) available from the U.S. Geological Survey. The matching accuracy of the stereo algorithms was evaluated by comparing the observed stereo disparity against the ground truth derived from the DEMs. It was observed that trinocular local matching reduced the percentage of mismatches having large disparity errors by more than half when compared to binocular matching. On the other hand, trinocular stereopsis increased the computational cost of local matching over binocular by only about one-fourth. We also present a quantization-error analysis of the depth reconstruction process for the nonparallel stereo-imaging geometry used in our experiments. I IntroductionWe present a quantitative analysis of the contribution of a third camera in improving the accuracy of local stereo matches, and its computational overhead in the stereo correspondence process. In the recent past, trinocular matching (Ayache & Lustman 1987;Ito & Ishii 1986;Peitikainen & Harwood 1986;Yachida et al. 1986) has been proposed as a new technique for achieving local matching in stereo analysis. Contemporary literature does not contain an in-depth analysis of the advantages gained in the local matching performance as a result of using trinocular instead of binocular stereo. Understanding the trade-off between the contribution of a third camera and the extra computations required for trinocular local matching is of significant interest for those involved in the design of stereo-based vision systems. We also present a quantization error analysis for the nonparallel stereo imaging geometry used in our experiments.Stereo analysis has been recognized as an important passive mechanism to extract 3D world information from intensity images. Earlier, Barnard and Fischler (1982) presented a detailed survey describing the major stages involved in stereo analysis and the different computational approaches developed starting from the mid70s up to 1981. The three main stages of computational stereopsis are preprocessing, establishing a correspondence, and computing the 3D depth (from the stereo disparity). Dhond and Aggarwal (1989) provide a comprehensive review of the major stereo algorithms developed in the recent past . Traditionally, only binocular (two view) images have been used in stereo analysis to...

Stereo correspondence based on line matching in Hough space using dynamic programming

“…4.80% for the corridor scene, and 2.98% for the office scene. The above dispanty errors compare very well with other reported result\ 1131- [15], [20]. For the cube images, all 72 pairs of lines are correctly matched, and the disparity errors occur mostly on the slanted edges in the cubes.…”

“…It is known that if the two cameras of a binocular system are displaced horizontally, then the horizontal lines in two images cannot be uniquely matched unless the end points of the horizontal lines can be located. A good solution to this is the use of three cameras as suggested by [13]- [15]. The trinocular stereo vision facilitates an additional powerful constraint, the trinocular uniqueness constraint [13], i.e., the horizontal and vertical disparities are identical, provided the camera displacements in both directions are equal.…”

Rough terrain profiling using digital image correlation

“…The disparity of a point can be mathematically shown to be inversely proportional to the depth of the point . It can easily be shown that if the geometry of the system, the orientation and translational position between the two cameras, as well as the camera intrinsic calibration parameters are known then the coordinate position of a point can be mathematically determined [13].…”

Novel high speed method using gray level vector modulation for 3D shape measurement

“…Although systems with two cameras and one projector are more complicated, they are also more flexible. One of its advantages is the reduction of matching ambiguity [4][5][6] . Since there are two cameras in the system, the lightingto-image correspondence is not necessary and only image-to-image stereo matching is enough for calculating depths [7] .…”