基于CS-SD的车载环境下实时行人检测模型

发布时间：2018-01-30 02:50

本文关键词： 行人检测方向梯度直方图行人区域特征提取车载环境　出处：《交通运输工程学报》2016年06期 　论文类型：期刊论文

【摘要】：针对车辆辅助驾驶系统中行人检测的实时性问题,提出一种基于路面边缘线标定结合显著性纹理检测(CS-SD)的算法和定位方向梯度直方图(L-HOG)的行人检测模型,应用CS-SD算法替代穷尽搜索快速标定图像中的行人区域,应用L-HOG快速提取行人特征,并采用附加核心的支持向量机(AK-SVM)进行高效目标分类。分析结果表明:在个人计算机上对包含832个行人的500幅图像进行检测时,模型正确检测720个行人,检测率为86.5%,误检率为4.1%,检测时间为39ms;在基于BF609的车载行人检测系统上对包含988个行人的48 400幅图像进行检测时,模型正确检测861个行人,漏检127个行人,误检13个行人,检测速度为20fps。可见,提出的行人检测模型在不降低检测率的前提下,可以达到满意的检测速度,并且可以用于实时行人检测车载设备。
[Abstract]:Aiming at the real-time problem of pedestrian detection in vehicle assisted driving system. A pedestrian detection model based on road edge line calibration and salient texture detection (CS-SD) and orientation gradient histogram (L-HOG) is proposed. The CS-SD algorithm is used to replace the exhaustive search for the pedestrian area in the fast calibration image, and the L-HOG is used to extract the pedestrian features quickly. An additional kernel support vector machine (AK-SVM) is used to classify the target efficiently. The analysis results show that 500 images including 832 pedestrians are detected on the personal computer. The model detected 720 pedestrians correctly, the detection rate was 86.5, the false detection rate was 4.1 and the detection time was 39 seconds. When 48,400 images of 988 pedestrians were detected on the vehicle pedestrian detection system based on BF609, the model correctly detected 861 pedestrians and missed 127 pedestrians. It can be seen that the proposed pedestrian detection model can achieve satisfactory detection speed without reducing the detection rate. And can be used for real-time pedestrian detection vehicle equipment.
【作者单位】：上海大学机电工程与自动化学院;山西轻工职业技术学院机电工程系;上海大学微电子研究与开发中心;
【基金】：国家自然科学基金项目(61376028)
【分类号】：TP391.41
【正文快照】： 0引言为了更好地保护路面上的行人并降低驾驶员的操作负担,行人检测逐渐成为车辆辅助驾驶系统中的一项关键技术。经过十几年的发展,行人检测技术取得了一定的成就[1-4]。方向梯度直方图(Histogram of Oriented Gradient,HOG)在行人检测中具有比较重要的意义[5]。在进行HOG特征

【相似文献】