Research On Integrated V2I And Vision Surveillance For Identification Of Container Truck Crash Risk In Terminal Area

The transportation in large container terminals is becoming increasingly intense,and visibility problems are common in the complex storage yard and tight exchange areas,which can lead to traffic crash between container trucks and facilities and operators.To enhance vehicle tracking and improve driving safety in intensive container terminals,under the Vehicle-to-Infrastructure environment(V2I),a method that integrate roadside camera and vehicle On-board Unit data to track the movement of trucks and then determine potential risk of traffic crash is proposed.The main research works of dissertation are as follows:(1)Established container terminal system structure by integrated V2 I and vision surveillance.Aiming at the limitations of the sight blocking of container truck drivers and the autonomous perception of single container truck,and the existence of non-CV container truck and operators without walkie-talkie in container terminal,the dissertation proposed a container truck crash risk identification method based on the integrated V2 I and vision surveillance.The machine vision algorithm can identify the target container truck and operators in the surveillance video,and then the roadside unit(RSU)can convert the target status into standard V2 I Basic Safety Messages(BSMs).The potential crash risk of container truck can be identified by integrated V2 I and vision surveillance,and the RSU can broadcast the warning to the container truck and operators within the communication range in real time.(2)Identifying container terminal target container trucks and operators.Firstly,the images of the container trucks and the operators are collected,and the image data are manually marked,generating the corresponding data set files.Secondly,YOLOv5 s model is used to train container trucks image data set,and YOLOv5 x model is improved by adding attention mechanism,so as to train container terminal operators’ image set.Then,the trained weight file is used to identify the container trucks and operators,generating the result of image object detection.Finally,the Image Pixel Coordinate-Global Positioning Coordinate method is applied to transform the target state into a standard BSMs of V2 I.(3)Integrating V2 I and vision surveillance data to identify the carsh risk of container trucks.Firstly,the Interactive Multi-Model(IMM)algorithm is applied to integrate the roadside vision surveillance and the location data of the On-board Unit(OBU),and real-time estimation and short-term prediction of the relative motion state of the container truck under maneuvering conditions are carried out.On this basis,the location of the container terminal target container trucks and operators is tracked,and the container truck crash risk is identified by setting the crash risk area model of container trucks.Finally,the effectiveness of the container trucks tracking and crash risk identification under the condition of data source integration is verified by setting an experimental scenario.The experimental results show that the proposed method of the dissertation is more conducive to achieving the container trucks maneuvering status tracking and crash risk identification,which in turn,identifies the container truck driving safety and caters the decision-making strategy for container trucks assisted driving system.