Research On Key Technologies In Driving Safety Assistance System

With the increasing demand for road vehicle safety,efficiency and intellectualization,as well as the breakthroughs in advanced technologies of communication,cloud computing,automatic control and computer vision,advanced driving safety assistant system and autonomous driving technology have become the key technical elements of Intelligent Transportation System(ITS).Therefore,this thesis focuses on vehicle driving safety assistant technology,which has significant theoretical significance and practical application value for improving vehicle active safety and road traffic environment.The works and contributions of this thesis are as follows.(1)Based on the preliminary investigation and analysis of the vehicle dispatching,we propose a dispatching method for vehicle safety assistance in terms of vehicle queues.Firstly,we establish the dispatching system model and propose a intelligent vehicle queuing method.Then we analyze the collision-free grouping of vehicle queues and the passing time of the intersection.Taking the throughput,safety and fairness of the vehicles into account,we set up the optimization objective of system and dispatching strategies.Finally,VISSIM and MATLAB are used to co-simulate,and the results verify that our proposed method can reasonably dispatch vehicles with the traffic flow of each road,and greatly improve the throughput of intersections.It can also alleviate traffic,and increase the road safety.(2)Based on the preliminary investigation and analysis of the warning methods for vehicle collision avoidance,a collision detecting method is present to monitor the specific accident areas.Firstly,we set up a system model.Then,we design a vehicle trajectory prediction network model based on Long Short-Term Memory(LSTM)module.The vehicle collision probability is calculated by the uncertain position relationship of vehicles.Moreover,a warning time parameter and threshold are set,which allow users to customize the sensitivity of collision warnings.Finally,the simulation results verify that our proposed method can not only effectively send collision warning to avoid risks,but also greatly reduce the system cost of computational and communication.(3)Based on the preliminary investigation and analysis of the traffic sign recognition,a dual multi-scale feature fusion method for traffic sign detection is proposed,which is designed on the basis of Single Shot MultiBox Detector(SSD)(4)network.Firstly,we study the characteristics of different layers' feature maps in deep neural networks.Then we design two different multi-scale feature map fusion structures based on SSD detection network,including adjacent feature fusion and pyramid feature fusion.Finally,a dual multi-scale feature fusion SSD structure is proposed.Finally,the proposed recognizing method is evaluated on the Chinese Traffic Sign Datasets(CTSD).The results verify that the proposed method has better overall performance in accuracy and speed,and can effectively detect traffic signs with different sizes in complex road scenarios.