Research And System Design Of Tarp Cover Recognition Method For Engineering Vehicles Based On Computer Vision

Target detection is not only a fundamental task in computer vision but also a prerequisite for performing other complex tasks related to computer vision.More and more target price detection algorithms appear in research and application fields.On the other hand,the national urbanization process is accelerating,and the demand for the transportation of construction debris and all kinds of engineering materials is growing.Engineering vehicles are overloaded and not covered with tarpaulin as required,causing severe traffic accidents and dust dumping problems are common.For the problem of many difficulties in managing the transportation of construction vehicles such as dumper trucks,this thesis helps traffic management departments to automatically identify construction vehicles passing through intersections and chokepoints and determine whether they are covered with tarpaulins by using the technology of target detection,to supervise the problems of overloading and dust spreading by construction vehicles.This thesis is based on the YOLO series algorithm,and the specific research contents are as follows.(1)For the high requirements of tarp recognition on recognition accuracy,based on the basic model of YOLOv5 s network,analyze the advantages of backbone module feature extraction,propose to add the mainstream three attention mechanisms SE,CBAM,and CA into Bottle Neck layer and C3 layer of C3,respectively,and explore the use of feature extraction capability of the model at different locations after attention to enhance The accuracy of engineering vehicle tarpaulin recognition using different attention mechanisms to improve the model is investigated.After separate training experiments,it is concluded that the model improvement of YOLOv5s-SE has the most considerable accuracy improvement compared with other methods,and the accuracy rate is improved by 0.08,and m AP is improved by 0.04 compared with the baseline model.(2)The research of YOLOv5 s target detection algorithm improvement based on multi-scale fusion is carried out for the problem of large-scale change of objects in complex scenes.It is proposed to use Bi FPN instead of PANet to enhance the feature map multi-scale fusion method.The experimental comparison yields better performance for the multi-target recognition task,with an accuracy improvement of 0.056 and an improvement of 0.017 in m AP compared to the original model.(3)Based on the benchmark model’s inadequate performance on this thesis’ s data set,a joint improvement method YOLOv5s-SE-Bi FPN detection algorithm is proposed to explore the impact between the joint improvement methods and the performance on engineering vehicle tarp cover recognition.The experimental results show that the recognition accuracy of the improved joint model on the self-built dataset is improved by 0.085,and the m AP is improved by 0.05 compared with the benchmark model.(4)Using LabelImg annotation tool to classify and annotate the data obtained from multiple channels,the development process of the engineering vehicle tarpaulin cover recognition system is explicitly described from four aspects: system analysis,system design,system presentation,and system testing,and the problems occurred in system testing are analyzed,and solutions are proposed.