一种基于多区域信息融合约束的改进帧间差分目标检测与跟踪算法

An improved interframe differential target detection and tracking algorithm based on multi-region information fusion constraints

针对帧间差分法在目标细节识别上较为粗糙的缺点,提出一种优化改进的目标检测与跟踪算法,构建两次区域限定与Kalman滤波算法融合的检测方法.首先通过提出的网格划分算法来获得动态感兴趣区域,仅在此道路的显著性范围内进行车道线区域的识别;然后依据车道线区域信息与建立的道路区域掩模作为帧间差分的评价指标,提高算法的处理精度,并使用扩展的Otsu算法,使用自适应动态阈值解决了帧间差分目标提取性能的有效性问题;最后基于检测的目标形心坐标作为观测量,通过Kalman滤波跟踪预测目标的行驶轨迹.采用Halcon视觉软件平台,对所提出算法性能进行试验验证.结果表明,所提出算法平均处理时间为32.132 ms,能够满足实时性需求,且车道线检测的平均准确率达95%以上.此算法能够迅速、准确地提取车道线区域,且对目标位置具有较高的可预测性.

To solve the problem of frame difference method with rough target detail recognition,an improved target detection and tracking algorithm was proposed,and a detection method combining two-time region restriction algorithm and Kalman filtering algorithm was constructed.The dynamic region of interest(ROI)was obtained by the proposed grid partitioning algorithm,and the lane line region was identified only within the salience range of the road.According to the lane line area information and with the established road area mask as evaluation index of the frame difference method,the processing accuracy of the algorithm was improved.An extended Otsu algorithm was adopted to solve the problem of effectiveness of target extraction performance with frame difference method by adaptive dynamic threshold.With the detected centroid coordinates of target as observation quantity,the target trajectory was tracked and predicted through Kalman filter.Halcon vision software platform was used to verify the performance of the proposed algorithm.The results show that the average processing time of the proposed algorithm is 32.132 ms,which can meet the real-time requirements.The average accuracy of lane line detection is more than 95%.The proposed algorithm can quickly and accurately extract the lane area and identify the target position in the environment.