Research On Object Detection Method For Intelligent Driving Based On Federated Learning

Intelligent driving is a safe and effective driving process that relies on computers and artificial intelligence methods to achieve a safe and effective driving process that is out of human control.With the advent of the era of artificial intelligence,intelligent driving has ushered in a new era in the field of driving.At the same time,the safety issues of smart driving have become more prominent.First of all,it will face the difficulty of processing large-scale data when cooperating with intelligent driving systems.Secondly,the accuracy of using object detection to judge road condition information during driving still needs to be improved.Finally,the data captured by the vehicle-mounted camera may reflect the location of the vehicle’s active area,thereby exposing private information.With the rapid development of deep learning,the current commonly used object detection algorithms are mostly based on deep learning.Although the detection effect based on deep learning algorithms is considerable,the centralized processing of largescale data will put huge pressure on the server and storage space,and will cause a lot of waste of computing resources.Moreover,since the current object detection model often directly uploads the original data,the privacy protection is poor.Federated learning is a machine learning paradigm suitable for large-scale distributed deep learning model training.It supports customers to use local data to train network models separately,and then aggregate global models together on a central server.Therefore,the combination of federated learning and object detection is of great significance for improving the safety and efficiency of intelligent transportation and protecting the privacy of vehicle users.In response to these problems,this paper evaluates the effects of existing object detection algorithms,and proposes an object detection method for intelligent driving based on federated learning for the first time.The specific work of this paper includes the following parts:(1)For object detection,this article focuses on some typical general object detection architectures,and uses a variety of detection methods to conduct experiments,so as to obtain the most suitable detection benchmark model for this article.(2)Traffic sign detection belongs to an application scenario of object detection.For traffic sign detection,this paper introduces a federated learning method.The collected image information is parameterized by using federated learning and then transferred to the central cloud for aggregation.This process avoids the occurrence of privacy leakage during data transmission,and the new model The average detection time for traffic signs increased by 0.15 s on average,which verified the important role of federated learning.This method completes the entire training process through three stages of data preprocessing,centralized pre-training,and federated training,which effectively protects the privacy of data transmission between vehicle users.(3)The method proposed in this paper is further improved,and the traffic sign detection dataset is divided into independent and identically distributed data.This paper uses the proposed method to conduct experiments on the German Traffic Sign Detection Benchmark(GTSDB)and the newly added Chinese Traffic Sign Dataset(CCTSDB).Experiments have proved the effectiveness of this method.The method proposed in this paper can improve the efficiency of object detection in smart driving cars,and has practical significance and practical value for the practicability and safety of smart driving cars in future life.