基于YOLO算法的行人检测方法

Pedestrian Detection Method Based on YOLO Algorithm

针对复杂道路交通环境,选择YOLO(You Only Look Once)实时目标检测算法,对行人目标进行检测识别的研究。YOLO算法在目标检测的速度和精度上都取得过良好效果。首先在YOLO网络模型的基础上针对行人单类检测问题,修改分类器,并通过卷积操作改变网络最后的输出维度;其次通过对道路交通场景下采集到的样本图片进行标注,得到行人数据集;然后采用相同预训练模型在YOLOv2和YOLOv3上训练,通过优化网络参数,加速模型收敛。实验结果分析可知,基于改进的YOLOv3的行人目标检测方法更能满足实时性的要求。

In view of the complex road traffic environment,the real-time target detection algorithm YOLO was selected to carry out the research of pedestrian target detection and recognition.YOLO algorithm has achieved good results in the speed and accuracy of target detection.Firstly,based on YOLO network model,the classifier is modified to solve the single-class detection problem of pedestrians,and the final output dimension of the network is changed by convolution operation.Secondly,the pedestrian data set is obtained by marking sample images collected under road traffic scene.Then,the same pre-training model is used to train on YOLOv2 and YOLOv3,and the network parameters are optimized to accelerate the model convergence.According to the analysis of experimental results,the improved pedestrian target detection method based on YOLOv3 can better meet the real-time requirements.