Research On Target Detection And Ranging Method In Front Of Mining Truck Based On Deep Learning

Mining trucks are the main transportation equipment in open pit mines,and safe driving of mining trucks has become a critical issue in open pit production.The cab of the mining truck is only a small part in front of the mining truck,so mining trucks have a visual blind spot.In order to solve the safety hazard caused by the visual blind zone of mining trucks,this paper proposes a method for detecting and measuring the front object of mining trucks based on deep learning,which is used to detect the type,location and distance of targets in mining truck blind areas.So as to realize the auxiliary driving technology for mining truck blind area monitoring.Since the common target detection category does not contain mining trucks,it is necessary to mark the pictures taken at the mine site and the pictures taken from the surveillance cameras,and combine them with the PASCAL VOC data set to create a self-built data set.In view of the problem of too few samples of mining trucks during model’s training,this paper uses data enhancement algorithms to expand the data of mining trucks.In this paper,tiny-yolov3 model is used to detect the target in the blind area of mining trucks.Since the self-built data set is used during training,this paper uses the improved K-Means clustering algorithm to calculate yolo anchor.The detection speed of tiny-yolov3 is fast but the detection accuracy is insufficient,and the target in the image taken in the mining truck environment is small,so in this paper,the residual structure is used to improve the tiny-yolov3 target detection model.In the improved model,the detection accuracy can be greatly improved when the detection speed is slightly decreased.When training the model,this paper uses step-by-step training to train the model with different batchsizes and learning rates,which can improve the detection accuracy of the model.For the picture taken in the mining truck environment,this paper uses the monocular vision ranging method under the overhead view to calculate the horizontal distance of the target.In this paper,the target detection model is combined with the monocular vision ranging algorithm,and a monocular vision ranging method based on feature points is proposed to measure the distance of targets in the blind zone in front of the mining truck.The method has a ranging error of less than 0.4 meters within 15 meters far from mining trucks,which basically meets the requirements of target distance measurement in the mining truck environment.According to the category of the target in the blind area in front of the mining truck,this paper uses different colors to label it,marks its position with a rectangular frame,and marks its category information and distance information above the rectangular frame to realize the auxiliary driving technology for mining truck blind area monitoring.