Research On Anti-collision Detection Of Unmanned Trucks In Open Pit Based On Machine Vision

Intelligence mine is the only way for the mineral industry to transform towards efficient and green development.The mining and transportation process of the mine has paid more attention to the integration of technologies such as artificial intelligence and big data.Open-pit mine unmanned truck transportation has become the focus of the development of unmanned mines in recent years due to its closed,low-speed and lowinteraction characteristics.The environmental perception of unmanned trucks is a prerequisite for driving decisions and motion control.Anti-collision detection is also a key link for vehicle travel perception.With the development of image detection technology,the image perception method based on machine vision has gradually become a hot spot for collision detection because of the advantages of low cost and abundant information.Based on the machine vision method,this article has made an in-depth study on the detection method of unmanned truck anti-collision in open pit mine.The main work of this paper is as follows:(1)The collision target detection method based on machine vision is studied.The high-precision Mask R-CNN network is selected as the basic detection network.For large and sparse targets in the process of carrying trucks in open-pit mines,combined with receptive field theory,dilated convolution is introduced in the residual module to expand the receptive field of the candidate detection frame and enhance the detection accuracy.Furthermore,the redundancy of the candidate box is analyzed,the candidate box generation method is improved to accelerate the network training to realize the highprecision collision target detection,which is a foundation for the subsequent research.(2)Based on the target detection results,the 5-dimensional distance estimation features of the detection frame width,height,center point coordinates and mask area are designed.Aiming at the situation where the difference in the proportion of different target pixels is large,a gradient boosting regression model is built separately and compared with the four distance estimation models in the public data set.Experimental results show that the gradient boosting regression model designed and constructed in this paper can effectively estimate the distance of large collision targets.(3)According to the distance estimation results,the LSTM distance prediction model is constructed to realize the overall process from input image to distance prediction,and a grid search process is designed for the hidden layer and time step selection problems,to further improve the accuracy of prediction.Experiments show that this paper proposes a anti-collision detection method for unmanned trucks in open-pit mines based on machine vision,which basically meets the accuracy requirements of unmanned truck collision detection in open-pit mines,and has certain applicability and feasibility.