基于视觉注意机制的行人目标检测

The Research of Pedestrian Detection Algorithm Based on Visual Attention

在分析现有运动目标检测算法的基础上,针对三帧差分检测易出现空洞及边缘不连续等问题,提出一种基于视觉注意机制的行人目标检测算法。该算法首先,通过借鉴人类视觉显著性现有的研究成果获得视觉显著图;其次,使用一种逻辑迭代三帧差分法,将其与基于区域搜索迭代的内轮廓填充法相结合,获得运动区域,融合两种算法提取出ROI;最后,采用HOG特征结合SVM分类器对ROI进行行人检测。对比实验结果表明,该算法连通性好,准确率高达97%以上。

Based on the analysis of the existing moving target detection algorithm, for the phenomena in three - frame difference detection, such as cavity and edge discontinuity ,, a pedestrian object detection algorithm based on visual attention mechanism was proposed. Firstly, we used the existing research results of human visual saliency to obtain the visual saliency map;Secondly, a logical iterative three-frame difference method was used to obtain the motion region combine with a contour filling method based on the local region research, two algorithms were integrated to obtain the region of interest. Finally, HOG features was used combined with support vector machines classifier for pedestrian detection in region of interest. The experimental results show that the algorithm has good connectivity and the accuracy is over 97 %.