Target Detection Of Large Trucks In Vehicle Environment Based On Machine Learning And Implementation Of Safety Auxiliary System

With the continuous improvement of mine safety production demand,the safety transportation of large mining trucks is becoming increasingly prominent,which becomes the key problem affecting the safety of production in mines.In order to solve the collision accident caused by the blind area of large truck vision,trucks panoramic auxiliary system shows a 360° image with the truck as the center.The auxiliary system can effectively eliminate the blind area of ground vision.However,due to the negligence of the driver,it is inevitable that the objects in the auxiliary system are not found and the accident will occur.To solve this problem,an improved SSD object detection model is proposed in this paper,and a large truck safety auxiliary system is implemented with Jetson TX2 platform.In this paper,a lightweight SSD object detection model based on atrous convolution is proposed to solve the problem of small object in the image of truck-mounted environment caused by the height of large trucks.The model solves this problem from two angles.Firstly,the redundant feature layer used to detect larger objects in the standard SSD object detection model is removed and a lightweight model is constructed.Secondly,atrous convolution layer is introduced,which is compatible with small object feature information and context semantics information,so as to improve the detection accuracy of small objects.Aiming at the problem of positive and negative samples imbalance in the process of generating candidate regions,this paper introduces objectness prior strategy to delete most of negative samples,and effectively balance positive and negative samples.The mine environment where large trucks are located is dull in color and the color information is redundant.In this paper,the gray image dataset is used instead of RGB dataset to train the model.Experimental results show that this optimization strategy improves the detection accuracy of the model.In order to solve the problems of insufficient image data and the difference between general dataset and self-collected dataset in large truck vehicle environment,this paper adopts the transfer learning strategy to fine-tune the parameters of the object detection model.The object detection model is pre-trained on the PASCAL VOC dataset,and then the pre-trained model is fine-tuned on the self-collected dataset to inherit the objectivity of the model.In order to reduce the storage space size of the object detection model and improve the detection speed of the model,the model is pruned and retrained for transplanting to embedded platform in this paper.Object detection model is transplanted to Linux operating system based on Jetson TX2 platform to realize large truck safety auxiliary system based on Jetson TX2 platform.The system uses industrial cameras to capture images of the truck-mounted environment and detect object in the Linux operating system.The system realizes image display through OpenCV library and marks the objects with bright color.It realizes safety warning of visual reminder by using prominent color.