关键词: 井下视频, YOLO, 目标边界框特征点, 图像消失点, 三角形模型, 不安全行为

Abstract: At present, the unsafe behavior of underground workers such as omission of inspection is one of the factors that affect the safety of coal mine enterprises. Aiming at the video data features of complex underground environment, single background and not obvious reference, it is difficult to measure the velocity of moving target directly by visual calculation. In order to solve the above problems, a speed detection method that does not rely on scene calibration is proposed, and the operation specifications of underground workers can be intelligently monitored through independent learning. The method is based on YOLOv3 to continuously detect and track the inspectors. In this process, the design uses the target boundary box feature points in the adjacent frames of the video to determine the image vanishing point; secondly, it uses the image vanishing point measurement real-time speed; finally, use the measured speed trend to intelligently judge whether underground workers have unsafe operation behaviors. The experimental results on coal mine data sets show that the proposed method can effectively measure the velocity of moving targets in the mine, and is used to judge whether there is missing detection behavior when underground workers patrol, which is more suitable for application in the underground scene than the calibration method.