Technology And Application Of Vehicle Video Object Detection And Driving Safety

At present,with the development of the Internet and artificial intelligence(AI),intelligent driving technology has attracted attention from all walks of life.In intelligent driving technology,the research of vehicle detection technology is particularly important.The accuracy and real-time of vehicle detection are related to the use of other modules of the intelligent driving system.The development of contemporary artificial intelligence has laid a solid foundation for the research of vehicle detection technology.The research in this article improves the accuracy of vehicle detection in complex scenes on the basis of object detection based on deep learning,improves the real-time vehicle detection capability of the algorithm and reduces the size of the occupied space in actual application scenarios.First of all,in response to the problem of the missed object for two-level detection based on deep learning in complex traffic scenes.This paper proposes a dual-channel vehicle video object detection algorithm.This method adds a depth information channel to supplement the depth details.Fully utilize the visual information of the two channels to describe the characteristics of the vehicles,improve the accuracy of vehicle detection and reduce the rate of missed objects.The method also adds a hard sample mining strategy to train the deep network model to solve the problem of the imbalance between negative samples and positive samples.The simulation results show that the method has obvious performance improvements in terms of algorithmic vehicle detection accuracy and missed detection rate.Secondly,in order to improve the real-time performance of the vehicle detection algorithm and reduce the memory occupied by the algorithm,this paper proposes a fast vehicle detection algorithm with deep network compression on the basis of the two-stage detection network.This method uses the scaling factor in the BN layer as a measure of channel pruning,and proposes a threshold adaptive strategy for channel pruning.The strategy introduces an upper limit of pruning ratio to avoid the problem of precision reduction caused by excessive pruning.The simulation results show that the method can effectively improve the algorithm’s operating efficiency,achieve the real-time video detection effect,and reduce the size of the model to achieve the purpose of running on mobile devices.Finally,based on the research of the algorithm’s accuracy and real-time,this paper designs a driving safety assistance system,which mainly implements the forward collision warning function.The system is mainly divided into vehicle terminal,data processing platform and warning system.The system uses the vehicle terminal to collect the image information and depth information of the front of the vehicle,and sends the information to the data processing center for data analysis.Finally,the warning system judges the processed data and promptly reminds the driver to avoid danger.