Haar特征提取IP核设计及前方车辆检测系统实现  

Design and implementation of extracting Haar feature IP in forward vehicle detection

作　　者：任亮朴 徐美华[2] 陈高攀[1] REN Liang-pu;XU Mei-hua;CHEN Gao-pan(Microelectronic R＆D Center,Shanghai University,Shanghai 200072,Chin;School of Mechatronic Engineering Automation,Shanghai University,Shanghai 200072,China)

机构地区：[1]上海大学微电子研究与开发中心,上海200072 [2]上海大学机电工程与自动化学院,上海200072

出　　处：《电子设计工程》2018年第14期107-111,116,共6页Electronic Design Engineering

基　　金：国家自然科学基金资助项目(61376028;61674100)

摘　　要：基于视觉的前方车辆检测是智能驾驶辅助系统中非常重要的组成部分,针对目前前方车辆检测中关于检测速度和精度的问题,采用部分积分图结构、流水线结构、乒乓操作、数据复用等方法提高Haar特征的计算速度;对提取到的Haar特征,采用Ada Boost进行降维操作,再用SVM(Support Vector Machine)进行分类训练,提升分类能力,设计完成了基于Haar+SVM前方车辆检测系统中的Haar特征提取的IP(Intellectual Property)核以及前方车辆检测的嵌入式系统。实验结果表明,检测精度可达到97.2%,系统处理速度平均为每秒21.6帧,验证了所设计的算法和系统的正确性和有效性。The front vehicle detection based on vision is a very important part of the intelligent driver assistance system,regarding the speed and accuracy of detection problems of vehicle detection,the integral graph structure,pipeline structure,ping-pong operation,data reuse and other methods to improve the calculation speed of Haar features;It improves the ability of classifier by using Ada Boost and SVM to process the extracted Haar features,a Haar feature extraction IP kernel and an embedded system for vehicle detection are designed and implemented based on Haar＋SVM front vehicle detection system.The experimental results show that the detection accuracy can reach 97.2%,and the system processing speed is average 21.6 frames per second. The correctness and effectiveness of the designed algorithm and system are verified.

关 键 词：车辆检测 HAAR特征 ADABOOST SVM 嵌入式系统 

分 类 号：TN47[电子电信—微电子学与固体电子学]