Vision based inter-vehicle distance estimation with extended outlier correspondence

In this paper, we propose a vision-based robust vehicle distance estimation algorithm that supports motorists to rapidly perceive relative distance of oncoming and passing vehicles thereby minimizing the risk of hazardous circumstances. And, as it is expected, the silhouettes of background stationary objects may appear in the motion scene, which pop-up due to motion of the camera, which is mounted on dashboard of the host vehicle. To avoid the effect of false positive detection of stationary objects and to determine the ego motion a new Morphological Strip Matching Algorithm and Recursive Stencil Mapping Algorithm(MSM-RSMA)is proposed. A new series of stencils are created where non-stationary objects are taken off after detecting stationary objects by applying a shape matching technique to each image strip pair. Then the vertical shift is estimated recursively with new stencils with identified stationary background objects. Finally, relative comparison of known templates are used to estimate the distance, which is further certified by value obtained for vertical shift. We apply analysis of relative dimensions of bounding box of the detected vehicle with relevant templates to calculate the relative distance. We prove that our method is capable of providing a comparatively fast distance estimation while keeping its robustness in different environments changes.