基于改进Faster R-CNN的行人检测算法

针对行人检测中复杂环境,提出一种改进Faster R-CNN的行人检测算法,使用深度卷积网络从图片中提取适合检测目标的特征。基于Faster R-CNN算法,以Soft-NMS算法代替传统NMS算法,加强Faster R-CNN算法对重叠区域的识别能力。同时,算法通过"Hot Anchors"代替均匀采样的锚点避免大量额外计算,提高检测效率。最后,将21分类问题的Faster R-CNN框架,修改成适用于行人检测的2分类检测框架。实验结果表明:改进Faster R-CNN的行人检测算法在VOC 2007行人数据集,检测效率和准确率分别提升33%、2.6%。

Pedestrian Detection Algorithm Based on Improved Faster R-CNN

Aiming at the problems of complex environment, an improved Faster R-CNN pedestrian detection algorithm is proposed. The deep convolution network is used to automatically extract the features that are most suitable for detecting objects from pictures. Based on the Faster R-CNN, the Soft-NMS algorithm is used to replace the traditional NMS algorithm, which enhances the ability of the Faster R-CNN algorithm to identify overlapping regions. At the same time, the algorithm replaces the uniformly sampled anchor points with "Hot Anchors" to avoid a lot of extra calculations and improves the detection efficiency. Finally, the Faster R-CNN framework for the 21 classification problem is modified into 2-class detection framework for pedestrian detection. The experimental results show that the improved Faster R-CNN pedestrian detection algorithm has increased the detection efficiency and accuracy in the VOC 2007 pedestrian data set, which increased by 33% and 2.6%, respectively.