A Research On Safe Driving Technology Of Intelligent Transportation System Based On Deep Learning

According to statistics from the World Health Organization,there are millions of injuries and deaths caused by traffic accidents worldwide each year,which shows the importance of vehicle driving safety.By introducing advanced technologies like artificial intelligence,communications etc.,Intelligent transportation systems can improve the efficiency and safety of the traffic system greatly.With the development of deep learning in recent years,there are more and more researches using deep learning technology for driving safety in intelligent transportation systems.Trajectory prediction based on deep learning is one of them.Trajectory abnormality detection and collision warning are of great significance to vehicle driving safety.However,current trajectory prediction algorithms are not sufficiently considered in time and space dimensions,resulting in relatively low accuracy of prediction.At the same time,the current solutions related to the application of deep learning to the driving safety have relatively large resource consumption for lightweight nodes such as vehicles,and these solutions are difficult to effectively deploy in actual systems.To address the two issues above,this thesis proposes a deep learning-based safe driving solution in intelligent transportation system.First,we propose a two-layer collaborative trajectory prediction algorithm based on deep learning.The algorithm uses the collaboration of the target vehicle and six nearby vehicles to perform short-term prediction.On the basis of short-term prediction,the collaboration between the target vehicle and the vehicles in the area are used in Long-term prediction.At the same time,this thesis designs a trajectory anomaly detection and collision warning scheme based on trajectory prediction to improve vehicle driving safety.The trajectory anomaly detection scheme compares the predicted trajectory with the actual trajectory to evaluate the degree of abnormality.The collision warning scheme predicts whether a collision may occur by rectangular collision detection at the sampling time point on the predicted trajectorys.Then,this thesis proposes the concept of multi-dimensional offloading,and design the offloading strategy for the safety driving scheme above with this idea.Multi-dimensional offloading refers to the offloading of the data,computing and function resources required by a computing task.We evaluate the proposed trajectory prediction algorithm by simulation and compare it with other algorithms.The simulation results show that the two-layer cooperative trajectory prediction algorithm proposed in this thesis has a better prediction accuracy.At the same time,based on the trajectory prediction algorithm,we simulate the abovementioned trajectory anomaly detection and collision warning scheme.The simulation results show that the scheme has a good anomaly detection rate and collision detection rate.Finally,we simulate the muti-offloading strategy of the safe driving scheme above,and compare it with no-offloading and computing-offloading.The simulation results show that the muti-offloading strategy can reduce the training and application time delay,and alleviate the data pressure of the vehicle.