复杂动态场景下基于道路平面提取的行人检测  

Pedestrian Detection Based on Road Surface Extraction in Dynamic Complex Scenes

作　　者：衡浩[1] 熊惠霖[1] 

机构地区：[1]上海交通大学自动化系,上海200240

出　　处：《计算机仿真》2013年第9期161-164,193,共5页Computer Simulation

基　　金：国家自然科学基金资助项目(61075106)

摘　　要：研究交通安全调度优化问题,提高交通安全的高级辅助驾驶系统中的行人保护技术,是近年来热门的研究方向。其中,最严峻的挑战是建立可靠的车载行人检测系统。与普通静态图像的行人检测相比,复杂动态场景下车载视频中的行人检测面临实时性要求高、光照条件复杂等难点。针对上述问题,提出结合HOG行人检测方法和道路平面提取技术,将目标搜索范围限定在道路平面区域。实验结果表明,改进方法在降低虚警数量和提高检测速度的同时,显著提高了对小目标行人的检测效率。The pedestrian protection system （PPS） used in the advanced driver assistance system （ADAS） to improve traffic safety has become a hot research area in scheduling and optimization of traffic safety.The major challenge of PPS is how to develop a reliable onboard pedestrian detection system.Compared to detecting pedestrian in static images,onboard pedestrian detection is facing some new difficulties,such as high real-time demand,wide range of illumination conditions and so on.To handle these challenges,an algorithm was proposed by combining the histogram of oriented gradient （HOG） features with road surface extraction technique,so that the search regions are only limited on the extracted road surface.Experiment result shows that this method can reduce the false alarm rate,improve the detection speed,as well as significantly improve the small pedestrian detection rate.

关 键 词：行人检测 路面提取 色彩空间 动态场景 梯度方向直方图 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]