融合运动检测与目标跟踪的目标检测方法  

Object detection method fusing motion detection and target tracking

作　　者：邬厚民[1] 吴卫祖[2] WU Hou-min 1,WU Wei-zu 2(1.School of Information Engineering,Guangzhou Vocational College of Technology and Business,Guangzhou 511442,China;2.School of Information,Guangdong Ocean University,Zhanjiang 524088,Chin)

机构地区：[1]广州科技贸易职业学院信息工程学院,广东广州511442 [2]广东海洋大学信息学院,广东湛江524088

出　　处：《计算机工程与设计》2018年第7期2084-2089,共6页Computer Engineering and Design

基　　金：2014年广东省优秀青年教师培养计划基金项目(YP2014001)

摘　　要：面向智能交通应用需求,提出一种行人和车辆目标检测方法。通过运动检测剔除复杂背景的干扰,缩小目标检测范围;采用结合方向梯度直方图和支持向量机构建的目标检测器,在运动区域图像块上检测目标,采用非极大值抑制策略进行目标筛选,剔除冗余目标检测结果;采用粒子滤波方法进行目标跟踪,将跟踪结果反馈给目标检测器,依据粒子的权重来修正目标检测器得到的检测得分。行人和车辆检测实验结果表明,该方法的真正率和处理帧率高,假正率低。A pedestrian and vehicle objects detection method for intelligent traffic application was proposed.The interference of complex background was eliminated and the range of object detection was reduced using motion detection method,and objects were detected on image patches of motion regions using the object detector constructed through histogram of oriented gradients and support vector machine,furthermore non-maximum suppression strategy was used to select objects and delete redundant results of object detection.Target tracking was executed using particle filter method,and the tracking result was fed back to the object detector for modifying the detection score according to the weight of the particles.The results of pedestrian and vehicle detection experiments show that,the proposed method has high true positive rate and detection frame rate,and low false positive rate.

关 键 词：目标检测 行人检测 车辆检测 运动检测 目标跟踪 粒子滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]