Localization Development Of Vehicle-mounted Intelligent Vision System Based On Loongson 2K

Facing the needs of informatized military operations,vehicle-mounted intelligent vision systems with situational awareness are an important part of ground combat vehicles.Therefore,the development of a vehicle-mounted intelligent vision system with reliable target detection,rich functional interfaces,and strong environmental adaptability is one of the important contents of the current research in the field of army ground equipment.This paper mainly develops a domestic vehicle-mounted intelligent vision system for ground combat vehicles based on Loongson 2K1000 processor.First of all,this paper studies the development status of vehicle-mounted intelligent vision systems for ground combat vehicles and the research status of target detection technology based on machine learning at home and abroad.Starting from the actual needs,a modular and standardized design concept is adopted to propose the overall design plan of the vehicle-mounted intelligent vision system.Secondly,a hardware platform for vehicle-mounted intelligent vision system is designed based on the Loongson 2K1000 processor.Through comparative analysis of domestic and foreign power conversion chips,a localized power module is designed based on the XC79618 HCC chip;a complex programmable logic device(CPLD)is designed by using the domestic HWD2210 chip to realize the expansion of the functional interface,which is convenient for later upgrades;and a system health monitoring is designed Module to realize real-time monitoring of system temperature,current and voltage to ensure safe and stable operation of the system.Third,a vehicle target detection algorithm is designed based on the improved YOLOv3 deep neural network.The calculation of the operation network is reduced and the real-time performance of target detection is improved by replacing the traditional convolutional network with a deep separable convolution to construct an anti-residual block model.The accuracy of target detection is improved by optimizing the boundary error of the target.The enhanced mixed data sets were produced based on the KITTI and USDC(Udacity Self-driving-car)data sets.Using the network negative feedback training strategy to optimize the training model and verify the algorithm.Finally,the system's hardware functions,performance,and environmental adaptability were tested by setting up a debugging environment.To verify the system's target detection function,the real road tests of traditional YOLOv3 algorithm and improved YOLOv3 algorithm was carried out respectively by simplifying the test platform.The test results show that the vehicle-mounted intelligent vision system designed in this paper has reliable hardware performance,strong environmental adaptability,and the real-time indicators and accuracy indicators of target detection,which meet the design requirements.