基于粒子群优化的复杂交通监控车辆检测与跟踪.docVIP

基于粒子群优化的复杂交通监控车辆检测与跟踪 　　摘 要： 城市交通监控视频的背景与前景变化均极为剧烈，导致交通监控对车辆的检测与统计准确率较低，对此，提出一种基于车辆空间移动特点与粒子像素聚类的车辆检测与跟踪方案。首先，基于高斯混合模型将权重与标准偏差比例较高的部分选为背景，由此实现前景区域的提取，同时，使用二值遮挡技术对提取的前景边缘进行优化处理；然后，提取前景区域的部分粒子，对粒子进行聚类处理，结合粒子的空间位置与移动向量来提高粒子的聚类准确率；最终，由于同一粒子簇可能为两个运动形式接近的多辆车组成，针对粒子簇的轴线等参数设置了限制条件，从而判断是否为同一车辆。对车辆的追踪则基于连续帧之间相同粒子簇的相似率比较实现。对比试验结果表明，该算法在剧烈变换的背景条件下具有较高的车辆检测准确率，错误率较低，优于其他同类型算法。 　　关键词： 城市道路交通； 高斯混合模型； 前景遮挡； 聚类； 车辆追踪 　　中图分类号： TN820.4?34； TP391.4 文献标识码： A 文章编号： 1004?373X（2016）14?0106?06 　　PSO?based vehicle detection and tracking in complex traffic monitoring 　　WANG Qingfen， SUN Xiuting， FAN Wei 　　（Department of Electrical Engineering， Shijiazhuang Tiedao University Sifang College， Shijiazhuang 051132， China） 　　Abstract： Since the traffic monitoring to vehicle detection and statistics has low accuracy due to the violent background and foreground variation of city traffic monitoring video， a vehicle detection and tracking scheme based on vehicle spatial displacement characteristic and particle pixel clustering is proposed. The sections with high proportion of weight to standard deviation are selected as the background on the basis of Gaussian mixture model to extract the foreground region， at the same time， the binary occlusion technology is used to optimize the foreground edge. And then， the part particles of the foreground region are extracted for clustering processing， and the clustering accuracy rate is improved in combination with the spatial position and motion vector of the particles. Since a particle cluster could be composed of vehicles with two similar motion modes， the limiting condition is set according to the particle cluster axis and other parameters to judge whether the particle cluster refers to the same vehicle. The vehicle was tracked based on the similar rate comparison of the same particle cluster between the continuous frames. The contrast test results prove that the proposed algorithm has high vehicle detection accuracy and low error rate in violent tr