Research And Application Of Pedestrian Detection Method In Coal Mine

In recent years,video surveillance technology has been widely used in the field of coal production.However,with the access of a large number of video data,it is impossible to rely solely on manual monitoring of the screen,which leads to the unsafe behavior of personnel,the unsafe state of objects,and the unsafe factors of the environment.The hidden dangers caused by unsafe factors cannot be found in time,which increases the risk of coal mine production.Therefore,it is of great significance to empower the existing video surveillance through the new generation of target detection technology,fully tap and utilize the existing surveillance video resources,and realize the intelligent perception of unsafe factors such as people,objects and environment during underground operations.However,due to the complexity of the coal mine environment,target detection still faces the following challenges :(1)Factors such as different pedestrian scales and large equipment occlusion in coal mines lead to missed and false detection of underground pedestrian detection;(2)Due to the limited computing resources and insufficient storage space of embedded mobile devices,the standard target detection model is difficult to deploy.In view of the above problems and challenges,this paper has carried out a series of studies.The main work and achievements are summarized as follows.(1)A network model based on MF-YOLOX-S is proposed to solve the problem of pedestrian detection in complex scenes such as coal mine equipment occlusion and different pedestrian scales.Firstly,the multi-scale attention module is filled into the Feature Pyramid Networks(Feature Pyramid Networks,FPN)to extract the multi-scale context information.Secondly,the feature enhancement module is used to increase the receptive field in FPN and enhance the representation ability of the original feature pyramid.Finally,the experimental results under the MS COCO dataset and the coal mine pedestrian dataset show that the average accuracy mAP of the proposed algorithm is improved by 1.96 % and 3.64 % respectively compared with the original YOLOX-S,which shows better detection ability in complex scenes of coal mines.(2)Aiming at the problem that the traditional target detection network YOLOX-S is difficult to deploy,a new lightweight network model of GD-YOLOX-S is proposed.Firstly,the deep separable convolution and Ghost module are used to replace the standard convolution in the YOLOX-S backbone network,so as to significantly reduce the number of parameters and computational cost,thereby improving the detection efficiency of the model.Secondly,Cluster-NMS is used to replace the standard NMS of the original network,and the detection accuracy of the model is further improved at the expense of small detection speed.Finally,the proposed model is evaluated on the VOC dataset.Experiments show that the number of parameters of the proposed model is 3.1 M,which is much lower than that of the original YOLOX-S.The average accuracy mAP and detection speed FPS reach 73.5 %and 71 FPS,respectively,which are higher than other lightweight models,verifying the effectiveness of the model.(3)Finally,based on the actual application requirements,a prototype system for counting coal mine personnel in and out of wells was developed.The system uses the B/S architecture model to build the system,uses the Django framework to encapsulate the proposed MF-YOLOX-S model,and uses LayUI+ SpringBoot technology to develop the Web system.The functions of pedestrian detection,pedestrian counting,violation alarm and attendance management are realized,which meet the needs of practical application of pedestrian detection in coal mine.The system uses the Docker container to achieve operating environment isolation and version control,and can be deployed in different operating systems and hardware environments.