基于改进YOLOv7的夜间行人检测算法

Night Pedestrian Detection Algorithm Based on Improved YOLOv7

针对夜间行人检测任务中存在的检测速度慢、漏检率高、黑夜场景下识别效果差等问题,提出一种改进YOLOv7的夜间行人检测算法。改进算法中,使用YOLOv7-tiny网络作为baseline,以满足准确率的同时兼具较高的检测速度,在网络head部分,使用CSP HorNet模块实现关键特征之间的高阶交互,并引入SimAM注意力机制,在不增加模型复杂度的情况下,使网络聚焦更多重要的特征信息。实验结果表明,改进算法在测试集上准确率(Precision,P)达到91.7%,召回率(Recall,R)达到81.4%,均值平均精度(mean Average Precision,mAP)提升2.9个百分点。改进算法在真实场景检测任务中,有效降低了漏检率及错检率,表现出良好的性能。

To address the problems of slow detection speed, high miss rate and poor recognition in dark scenes existing for the night-time pedestrian detection task, we propose an improved night pedestrian detection algorithm based on YOLOv7.For the improved algorithm, the YOLOv7-tiny network is used as the baseline to meet the accuracy rate while combines high detection speed. In the head part of the network, the CSP HorNet module is used to achieve higher-order interactions between key features, and the SimAM is introduced to focus the network with more important feature information without increasing the complexity of the model.The experimental results show that the improved algorithm achieves 91.7% Precision(P), 81.4%Recall(R) and 2.9 percentage points improvement in mean Average Precision(mAP) on the test set. The improved algorithm effectively reduces the miss detection rate and error detection rate in the real scene detection task, which shows a good performance.