煤矿井下巷道变形巡检视频异常检测方法

Anomaly detection method of inspection video for coal mine underground roadway deformation

采用智能视频巡检技术进行煤矿井下巷道变形检测时,常用的背景差分算法因要求输入图像具有良好的时空连续性而无法满足巡检视频背景建模要求。根据煤矿井下巷道变形巡检机器人匀速、定向运动及周期性采集视频数据的特点,提出一种巡检视频异常检测方法:结合巡检机器人定位信息对巡检视频分段并提取相应关键帧,采用均值哈希算法建立背景模型,对背景模型中图像进行特征跟踪以实现校正,之后将背景模型与关键帧进行差分运算,生成二值掩膜并进行去噪及连通处理后,输出异常检测结果并更新关键帧。实验结果表明,该方法在一定条件下可较准确地定位关键帧并检测出异常目标,检测速度约为50帧/s。

When using intelligent video inspection technology for underground coal mine underground roadway deformation detection,the commonly used background difference algorithm cannot meet the requirements of inspection video background modeling due to the requirement of the input images having good temporal and spatial continuity.According to the characteristics of uniform speed,directional movement and periodic acquisition of video data of the deformation inspection robot in underground coal mine,an inspection video anomaly detection method is proposed.The method segments the inspection video with the inspection robot positioning information and extracts the corresponding key frames.Then the method establishes a background model based on the mean hash algorithm,and performs feature tracking on the frames in the background model to obtain correction.The method carries out a difference operation between the background model and the key frames to generate a binary mask and perform denoising and closed computing processing.Finally,the anomaly detection results are output and the key frames are updated.The experimental results show that the method can locate key frames and detect abnormal targets accurately under certain conditions,and the detection speed reaches about 50 frames/s.