Research And Implementation Of Safety Detection System Of Connected Vehicle Based On Distributed Deep Learning

With the explosive growth of big data and the improvement in computers'performance,artificial intelligence is playing an increasingly important role in many fields.Among them,deep learning technology is a crucial component of artificial intelligence.Using neural network,deep learning has achieved excellent results in the fields of image recognition and processing.The visual navigation of the self-driving vehicle is an important application of target detection.By detecting objects such as pedestrians,vehicles and obstacles on the road,self-driving vehicles can automatically correct the driving route and avoid danger.Some intelligent hardware on the market satisfy this demand,but they are expensive and have complicated installation steps.On the other hand,smartphones' computing capability is strong,and they are portable and easy to operate.Therefore,smartphone applications are gradually becoming more and more popular.At present,the mainstream smartphone-based object detection and identification methods are mainly divided into two categories.One is the real-time acquisition of images by mobile devices,followed by the implementation of deep learning tasks relying on cloud computing,and then the processed results are delivered back to mobile devices.Although such methods are accurate,they cannot be performed in the event of poor network conditions.The other is to use the mobile device's CPU and GPU to process the image,but the power and memory overhead can severely limit the performance of the calculation.In order to solve the above problems,this paper proposes a target detection and recognition system based on mobile devices and cloud for the intelligent traffic scenarios using deep learning.The system compresses the neural network model by low rank decomposition,parameter pruning and so on.The mobile devices use the model scheduling algorithm based on dynamic programming to maximize the accuracy of target detection under the restriction of the memory,power consumption and network bandwidth of the mobile device.This paper proposes a method to expand the training set by using semi-supervised learning.By integrating pictures with high confidence in results into the training set,the generalization ability of the model can be continuously improved.In addition,this paper also proposes a collision forewarning system based on smartphones' bluetooth low energy in terms of car and vehicle communications,which effectively improves safety.Compared with smartphone system,the safety detection system based on mobile cloud computing deep learning reduces memory overhead and saves power.Compared with the cloud computing system,the task execution state can be maintained even without network connection which enhanced stability and is of great importance to the application and development of mobile internet and self-driving vehicles.