Moving Object Distance Estimation Method Based on Target Extraction with a Stereo Camera

摘要

Real-time distance estimation with high accuracy is crucial to robot applications such as navigation and obstacle negotiation. However, considering the truth that high accuracy brings huge computation, the previous work either accelerates the distance estimation process through hardware, or only estimates the static objects, or sacrifices accuracy to realize real-time. In order to improve the trade-off between speed and accuracy, in this paper we propose a stereovision-based method that integrates the object detection and tracking into stereo matching. We utilize kernelized correlation filter (KCF) to track the moving object, and then extract target regions from image pair as input images to compute disparity map. In addition, we update each pixel's disparity by a simple but effective approach to further refine disparity map. Eventually, distance is calculated with disparity by triangulation. Experimental results show that our proposed method is suitable for dynamic environment, the extraction method improves the speed of the entire system, and the average accuracy of distance estimation can achieve over 95% from 88 cm to 300 cm when baseline is 151 mm.