基于双目视觉的车辆检测及测距

Vehicle Detection and Inter-Vehicle Distance Estimation Based on Binocular Vision

汽车测距系统在驾驶辅助系统中越来越重要,基于视觉的测距系统成本低实现简单,但精度易受算法本身的影响。提出一种基于双目视觉的实时车辆检测和车距计算的算法,算法利用类Haar特征和AdaBoost算法训练分类器进行车辆检测,提取车辆候选区域。同时,该算法提出一种基于双目系统的交叉再检测的方法降低误检。立体匹配算法方面采用一种由粗到精的匹配策略,提高双目测距算法的精度和性能。实验结果表明该方法具有精度高、鲁棒性强的优点。

Ranging system on vehicle is widely used in driving assistance system.Vision-based ranging system has the advantages of lower cost and easier implementation.However,its accuracy depends on the algorithm itself.Proposes a real-time vehicle detection and inter-vehicle distance estimation algorithm based on binocular vision system.The method uses Haar-like features and AdaBoost algorithm to run a classifier,by which we can do vehicle detection and extract vehicle candidate areas.Meanwhile,uses a crossover re-detection method based on binocular vision to reduce false detection in the algorithm.And adopts a coarse-to-fine stereo matching scheme to improve the accuracy and time performance of binocular distance estimation algorithm.Experimental results show the high accuracy and robustness of the proposed method.