Research On Auxiliary Driving Of Mine Electric Locomotive Based On Machine Vision

As one of the main ways of auxiliary transportation in mines,the monotonous working state,complex lighting conditions and unevenly distributed oxygen content in roadway of electric locomotive are easy to cause dangerous error of judgment and fatigue of drivers,resulting in serious safety accidents.In view of this,the research of auxiliary driving of mine electric locomotive based on machine vision is carried out,and the realtime detection of obstacles in front of electric locomotive running and the fatigue state of electric locomotive driver is realized on the edge embedded processor.The main research contents of this thesis are as follows:(1)Research on lightweight improved method of YOLOv5 s target detection.Aiming at the situation that the current deep convolutional neural network cannot meet the realtime requirements on the edge embedded processor,the lightweight network ShuffleNetv2 is used to optimize the YOLOv5 s algorithm.At the same time,Mish activation function and lightweight ECA attention mechanism were introduced into ShuffleUnit to integrate global feature information,which enhanced feature extraction ability and improved network detection accuracy,and solved the problem that ShuffleNetv2,as a backbone network,was fast and slightly poor in accuracy.Finally,S2＿MECA-Yolo V5 S can meet the real-time requirements and ensure the detection accuracy on the vehicle embedded processor.(2)Research on obstacle detection method in track dangerous area.After analyzing the actual transportation situation of mine electric locomotive,an obstacle detection algorithm for track dangerous area is proposed,which uses edge detection operator and curve fitting algorithm to divide the track dangerous area in real time,and combines the object detection algorithm to comprehensively judge the obstacle.The track obstacle detection experiments are carried out on the edge embedded processor.The results show that the ROI-based obstacle detection algorithm has good performance in obstacle detection.(3)Research on fatigue driving detection method based on facial features.In order to eliminate fatigue driving of electric locomotive drivers,firstly,lightweight YOLOv5 s was used to optimize the KCF face tracking algorithm,which solved the problem of face tracking loss and realized the continuous tracking and detection of face targets.The PFLD algorithm is used to extract the facial key feature points,and the fatigue detection of the electric locomotive driver is carried out by combining the fatigue discrimination standard and the discrimination threshold.The experimental results show that the fatigue detection algorithm proposed in this thesis has good detection performance and can accurately and continuously carry out real-time fatigue detection.The research on safety auxiliary driving method of mine electric locomotive carried out in this thesis solves the problem that the embedded processor on board cannot run the neural network algorithm,realizes the accurate detection of obstacles in front of the electric locomotive and the fatigue state of the electric locomotive driver,and provides a new idea for the intelligent construction of safe driving of mine rail transportation.Figure [43] Table [12] Reference [83]...