基于行人与车辆关系模型的行人检测

PEDESTRIAN DETECTION BASED ON PEDESTRIAN AND VEHICLE RELATION MODEL

针对在交通场景下的行人,考虑到绝大多数交通场景中车辆与行人同时存在,提出一种在对背景图像进行初步行人检测的同时对车辆进行检测的方法,建立一种行人与车辆关系模型。以车辆位置作为辅助检测基础,引入真假阳性检验用以排除出现在不可能区域的行人并介绍了具体方法。该方法首先对行人、车辆、行人与车辆关系特征进行定义并建模形成与其有关的函数关系,然后推导得到适用于支持向量机的标准形式,最后通过支持向量机回归法训练分类器进行分类识别。现场实测结果表明,此种方法大大降低了误检率,对不同分辨率图片中的行人均有较好的识别效果。

F o r the pedestrians in tra ffic scenarios, and considering th a t the vehicles and pedestrians coexist in m a jo rity o f tra ffic scenes,we p u t forw ard a m ethod th a t detects vehicles on background image w h ile carries outs p re lim in a ry pedestrian d e te ctio n , and b u ild a pedestriansand vehicles re la tio n sh ip m odel. W e use the locatio n o f veh icle as the basis o f a u x ilia ry d e te ctio n , introduce the true and false po sitive tests toexclude the pedestrians appearing in im possible re g io n , and present the spe cific m ethod. T h is m ethod firs t defines p e destria ns, v e h ic le s , andthe characteristics o f pedestrians and vehicles re la tio n s h ip , and m odels them to form the correlated fu n c tio n re la tio n . T h en th is is deduced toob tain the standard form suitable fo r support vector m achine. F in a lly the m ethod uses the support vector m achine regression m ethod to tra inthe c la ssifie r fo r c la ssifica tio n and recognition. Site measurem ent results show that this m ethod greatly reduces the rate o f error d e te ctio n , andhas good re cognitio n effect on pedestrians on the pictu res w ith d iffe re n t resolutions.