基于多线激光雷达与视觉信息的实时行人检测  

Real-time Pedestrian Detection Based on Integration of Multi-line Radar and Visual Information

作　　者：陆怡悦 蔡云飞[1] 石庭敏 

机构地区：[1]南京理工大学计算机科学与技术学院,南京210094

出　　处：《计算机与数字工程》2017年第11期2256-2262,共7页Computer & Digital Engineering

基　　金：国家自然基金青年项目(编号:61305134);博士点基金(编号:20133219120035);核高基重大专项(编号:2015zx01041101)资助

摘　　要：行人是无人自主车辆的重要感知内容。针对传统行人检测方法中检测速度慢的情况,提出了一种融合多线激光雷达和视觉信息的实时行人检测方法。该方法通过对激光雷达和摄像头的联合标定,将雷达的坐标映射到图像坐标中。根据雷达的栅格地图检测凸障碍,由此获得图像中的感兴趣区域。最后利用梯度方向直方图(HOG)和支持向量机(SVM)对感兴趣区域进行行人检测。实验验证了该方法显著地提高了检测速度和准确度,并弥补了单一传感器在行人检测中的不足。该方法在车辆以30km/h左右的速度行驶,环境背景不断变换的情况下都能很好地实现,具有实际应用价值。Pedestrians are the important aware contents of the unmanned autonomous vehicles. Since the slow detection speedof the traditional pedestrian detection methods ,a real-time pedestrian detection method based on integration of multi-line laser ra.dar and visual information is proposed. Firstly,combine laser radar with the camera by calibration,then map the radar coordinatesto the image coordinates. Secondly,according to the raster map of radar,projections obstacles are detected,thereby obtaining theregion of interest of an image. Finally,the gradient direction histogram（HOG）and support vector machine（SVM）are used to de.tect pedestrians of the regions of interest. Experimental verification of the method significantly improves the detection speed and ac.curacy and makes up for the lack of a single sensor in Pedestrian Detection. This method can be well implemented in the vehicle at aspeed of 30km / h or so of driving. At the same time,environmental background constantly changes. thus this method has practicalvalue.

关 键 词：联合标定 感兴趣区域 梯度方向直方图 行人检测 

分 类 号：TN95[电子电信—信号与信息处理]