基于电子稳像的嵌入式行人检测系统  

Embedded pedestrian detection system based on electronic image stabilization

作　　者：杨飚[1] 魏雪巍 YANG Biao;WEI Xue-wei(Beijing Key Laboratory of Urban Road Traffic Intelligent Control Technology,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学城市道路交通智能控制技术北京市重点实验室,北京100144

出　　处：《传感器与微系统》2018年第7期99-101,共3页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61374191);北京市教育委员会科技计划资助项目(KM201710009001)

摘　　要：设计并实现了一种基于电子稳像处理的数字信号处理器(DSP)嵌入式平台的行人检测算法:采用平滑特征轨迹法对输入视频进行电子稳像处理;采用sobel边缘算子提取人体的头肩边缘图像,根据改进Hausdorff相似性度量原理,提出了基于改进Hausdorff距离头、肩边缘模板匹配的行人目标检测算法;通过卡尔曼滤波算法对行人目标进行实时跟踪。实际路口测试结果表明:在TMS320DM8168嵌入式平台上设计与实现的基于改进Hausdorff距离头肩边缘模板匹配的行人检测算法可以对行人实现实时检测和实时跟踪,结合电子稳像算法,可以达到95%的检测率、9帧/s的检测速度,而误检率为4%,能够满足实际使用需求。A new pedestrian detection algorithm based on digital signal processor（DSP） embedded platform based on electronic image stabilization processing is designed. The smoothing feature trajectory method is used to deal with the electronic image stabilization processing on input video. The Sobel edge operator is used to extract the image of the head and shoulder edge of the human body. According to the improved Hausdorff similarity measurement principle, a pedestrian target detection algorithm based on improved Hausdorff distance head- shoulder edge template matching is proposed. The pedestrian target is tracked by Kalman filtering algorithm in reahime. The actual road test results show that pedestrian detection algorithm based on improved Hausdorff distance head-shoulder edge template matching designed and implemented on the TMS320DM8168 embedded platform can achieve real-time detection and real-time tracking for pedestrians, detection rate of 95 % , the detection speed is 9 frames per second, and error detection rate of 4 % can be achieved, it can meet the demand for actual use.

关 键 词：行人检测 TMS320DM8168 电子稳像 改进HAUSDORFF距离 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]