Design Of Construction Vehicles Action Recognition System Based On Object Detection

In recent years,the safety of oil and gas pipelines in high consequence areas(HCA’S)has attracted great attention in our country.At present,the inspection methods of pipelines in China are mainly based on manual inspections and supplemented by intelligent inspections.Because of the frequent construction near the pipeline and the destructive effects of these construction vehicles,it is necessary to obtain information on the categories and actions of various construction vehicles with fixed monitoring or inspection drones during the intelligent inspection process,so as to further improves pipeline safety.In this context,an intelligent recognition system of construction vehicles,which can realize the functions of motion detection,object detection,object tracking and action recognition of various construction vehicles in construction scenes,is designed and realized in this dissertation.The main works are summarized as the following three aspects:(1)The overall design scheme of the action recognition system of construction vehicles is formulate.By analyzing the functional requirements and performance indicators of the system,an overall design scheme is formulated for the system,and the motion detection module,object detection module,object tracking module and action recognition module in the scheme are demonstrated,respectively.(2)The action recognition system of the construction vehicles is realized.First,the dualmode single Gaussian model is used to establish the motion background model of the video to complete the motion detection.Secondly,the YOLOv5 algorithm is adopted to detect object on the video to obtain the category and location information of the construction vehicle.Thirdly,the simple online and realtime tracking(SORT)algorithm is used to match the construction vehicle between the adjacent frames to obtain the trajectory of each construction vehicle.Finally,the video spatiotemporal model of the matched construction vehicles is established to compute the specific action of each construction vehicle.After the algorithms of the function modules of the system are completed,the various function modules of the system are integrated.And the interface is realized by using PyQt.The system can intelligently recognize the construction vehicles through four different modes,and the recognition results can be displayed intuitively in the interface.(3)The action recognition system of construction vehicles is debugged and tested.The test results show that the action recognition system of construction vehicles based on object detection,which is designed in this dissertation,can meet the expected requirements and improve the efficiency of pipeline inspection.