Research On Collision Detection And Early Warning Of Shared Electric Vehicle Based On Deep Learning

In recent years,under the turmoil of severe air pollution and energy shortages,electric vehicles have gradually gained recognition from governments and people worldwide due to their near zero emissions,high energy efficiency,and low noise.However,the high price,limited recharge mileage and inconvenience of charging have become major obstacles to the large-scale promotion of electric vehicles.The use of sharing strategy to promote electric vehicles to society has become an effective method to deal with this problem.The shared electric vehicle can solve the problems of high purchase prices and immature technology,and meanwhile can effectively relieve traffic congestion and improve the urban environment.Therefore,shared electric vehicle not only meet the needs of users,but also meet the demands of economic development and environmental protection.A major issue in the electric vehicle sharing is to analyze the behavior of users,especially whether there are any dangerous driving behaviors such as scratches,collisions or potential collisions during the driving process.This paper adopts image object detection and segmentation algorithm based on deep learning to detect vehicles’ position,and proposes a vehicle trajectory prediction method based on video prediction and bounding box prediction algorithm.The main research content of this paper is as follows:1)The background and status quo of the development of shared electric vehicles are introduced,and the current research status in the field of automobile collision detection and early warning is summarized.In addition,the basic theories of deep learning related to this topic are introduced,the latest research progress of deep learning is presented;2)After researching and comparing various object detection algorithms based on deep learning,a high-precision,real-time object detection model—Single Shot Multibox Detector—is selected as the basic model for object detection.In addition,experiments are conducted on the vehicle detection data which tagged for this project.In order to further improve the operational efficiency of the object detection model,the feature extraction part of the SSD model is optimized,the classification and positioning layer were tailored,and a quantized algorithm was used to further compress the model.Eventually,experimental comparison was made between the optimized SSD model and previous SSD model.What’s more,the pixels-distance transform model is proposed,combined with the vehicle detection model,the distance of car could be calculated.3)The image segmentation algorithms are briefly introduced,and a model based on the skip connection and astrous spatial pymarid pooling —SkipASPPNet— is proposed,experiment has been done to verify the effectiveness of the model.In addition,the feasibility study of collision detection and early warning based on image segementation has been done.4)Mainstream video prediction algorithms are introduced.Video prediction algorithm based on convolutional LSTM and Geneative Advisial Network(GAN)is proposed to predict the vehicles’ tracjectory,thus can infer the position of the vehicles in adance and make early warning about potential collision.In order to improve the accuracy of trajectory prediction,bounding boxes’ tracjectory prediction model based on LSTM is proposed and the experiment is researched.