基于改进的YOLOv3行人检测方法

Pedestrian Detection Method Based on Improved YOLOv3

本文通过研究行人检测中普遍存在的目标遮挡、目标尺寸过小以及目标背景复杂等问题对目标检测算法性能所造成的影响,在YOLOv3行人检测算法的基础上从分类器、聚类方法、损失函数等角度出发,对YOLOv3算法进行改进。本文将改进后的YOLOv3算法与SSD算法、Faster R-CNN算法进行检测对比,试验结果表明,改进的YOLOv3算法在检测性能上得到了明显提升。

This paper studies the impact of the common problems in pedestrian detection,such as target occlusion,small target size and complex target background,which impact the performance of object detection algorithm,based on the YOLOv3 pedestrian detection algorithm,the YOLOv3 algorithm is improved from classifier,clustering methods,loss function,etc.At the same time,the paper compared the improved YOLOv3 algorithm with the SSD algorithm and the Faster R-CNN algorithm,the experimental results show that,the improved YOLOv3 algorithm has significantly improved the detection performance.