Efficient disparity calculation based on stereo vision with ground obstacle assumption

This paper presents a fast local disparity calculation algorithm on calibrated stereo images for automotive applications. By utilizing the ground obstacle assumption for a typical road scene, only a small fraction of disparity space is required to be visited in order to find a disparity map. It works by using the neighbourhood disparities of the pixels in the lower image line as supporting points to determine the search range of its upper vicinity line. Unlike the conventional seed growing based algorithms that are only capable of producing a semi-dense disparity map, the proposed algorithm utilises information provided by each pixel rather than trusting only the featured seeds. Hence, it is capable of providing a denser disparity output with low errors in homogeneous areas. The experimental results are also compared to a normal exhaustive search (block matching) algorithm, showing a factor of ten improvement in speed, whilst the accuracy is enhanced by 20% without constraint to the maximum possible disparity.