改进的YOLOv3红外视频图像行人检测算法

An improved infrared video image pedestrian detection algorithm

针对YOLOv3检测红外视频图像行人时存在准确率低、漏检率高的问题,提出一种改进的YOLOv3红外视频图像行人检测算法。根据行人在红外图像中呈现宽高比相对固定的特点,利用k-means聚类方法选取目标候选框个数和宽高比维度,调整网络参数并提高输入图像分辨率,最后进行多尺度训练得到最优检测模型,从而检测红外视频图像序列中的行人目标,并通过候选框标注行人位置。在CVC-09红外行人数据集上进行对比实验,结果表明,改进的YOLOv3算法在红外行人检测中的准确率高达90.63%,明显优于Faster-rcnn和YOLOv3算法,且改进后的网络能够同时检测到更多目标,降低了漏检率。

YOLOv3 detecting on infrared video images of pedestrians has low accuracy and high missed detection rate. An improved YOLOv3 infrared video image pedestrian detection algorithm is therefore proposed. According to the characteristics that pedestrians showed are relatively wide aspect ratio in infrared images, k-means clustering method is used to select the number of target candidate frames and aspect ratio dimensions, to adjust network parameters and improve input image resolution, and finally to perform multi-scale training. The optimal detection model is obtained to detect the pedestrian target in the infrared video image sequence and to label the pedestrian position through the candidate frame. The comparison experiments on the CVC-09 infrared pedestrian dataset show that the improved YOLOv3 algorithm has an accuracy of 90.63% in infrared pedestrian detection, which is significantly better than the Faster-rcnn and YOLOv3 algorithms, and the improved network can detect more targets at the same time, and therefore reduce the rate of missed detection.