Depth hole filling and optimizing method based on binocular parallax image

Environment perception is one of the most vital function for autonomous robots while performing complex tasks in dynamic environment. In binocular stereo vision algorithm, the calculation of disparity image depends on matching algorithm, and the unmatched points form depth holes in the image. When encountering obstacles with less texture, such as blank walls, large area deep holes will often appear. Although the existing filtering algorithm can solve the problem of small area deep hole, it cannot be applied to large area deep hole. To solve this problem, a depth hole optimization algorithm is proposed, which detects large area depth holes from the image globally, fills the depth holes according to the disparity value distribution, and optimizes the weighted least squares filtering effect. The experimental results show that the average time of the algorithm is only 15 ms.