改进 YOLO 算法的行人检测和慢行规划应用

Application of Improved YOLO Algorithm for Pedestrian Detection and Slow Traffoc Planning

摘要:

行人是物体检测应用中非常重要的一类目标，在检测行人的同时对人流数量进行实时精准统计在诸多领域有着很大的实用意义。通过搭建深度学习环境，设定行人数据集并对其进行重新标注、训练和测试后，对统计视频中行人目标的正确率、漏检率和误检率行了统计。结果表明：基于改进YOLO v4的行人检测模型，能够更加准确、高效地识别密集场景下的行人目标，达到了预期目标，从而能够为城市慢行系统规划和商业区规划等领域提供应用价值。

Abstract:

Pedestrians are a crucial category of targets in object detection applications, and the real-time and accurate counting of pedestrian flow has the significant practical significance in various fields. By setting up a deep learning environment, establishing the pedestrian datasets, and re-annotating, training and testing the pedestrians, the accuracy, missed detection rate and false detection rate of pedestrian targets in videos are statistically analyzed. The results indicate that the pedestrian detection model based on improved YOLO v4 can more accurately and efficiently identify the pedestrian targets in dense scenes, and the expected goals are achieved so as to provide the application value for planning of urban slow-traffic system and commercial districts.