Research On Front Vehicle Recognition Algorithm Based On Embedded ARM Chip

Improving the recognition rate of vehicles in front is the key technology for the development of advanced driving assistance systems(ADAS),which is of great significance to the development of active safety technology and smart vehicle industry.This project draws on machine learning theory in the field of artificial intelligence and edge computing theory in the field of IoT,combines Haar feature classifier training technology and Adaboost machine algorithm,and uses the computer vision platform OpenCV to perform off-line training of classifiers.Subsequently,an online recognition system based on edge computing was built jointly with the 4G network module.The vehicle classifier file was loaded into the OpenMV module of the embedded machine vision used,then combined with image acquisition and processing technology,research was conducted on the online detection of vehicles in front.Finally,with reference to relevant national standards,a test plan was developed and a real vehicle road scene test was conducted.It has been verified that the requirements for vehicle target identification in FCW and AEB systems are basically satisfied,and the designed vehicle identification system and method have certain feasibility and practicability.This paper is based on the sub-project of the Collaborative Innovation Center for Passenger Cars and Special Vehicles in Fujian Province-"Development of Driving Safety Aid System Based on Laser Ranging and Visual Information Fusion".The main research contents include:Vehicle classifier's off-line training and generation based on OpenCV.It mainly discusses the training and generation of the classifier files used in the vehicle identification,including the vehicle feature extraction based on Haar and the training of the weak classifier,and the cascade classifier generation based on the Adaboost machine learning algorithm.Finally,the software test of the classifier file under the sample image of different road scenes is also carried out on the VS platform.The failure scenarios of the test results are also analyzed,and the improvement ideas and suggestions are given.OpenMV online identification based on OpenMV.This paper studies the machine vision OpenMV module of STM32H7 series single core MCU as the main control chip,explores the development and application of MicroPython language in embedded application layer program,and improves its image acquisition and processing technology.On the other hand,the firmware of OpenMV has been redeveloped and explored the cross-compilation technology for ARM chips based on Linux.In addition,the process of online recognition based on the edge computing concept,the loading and testing of the classifier file on the OpenMV module,and the debugging and optimization of the embedded software program are also discussed in detail to realize the online real-time recognition function of the vehicle ahead.Road scene verification and real vehicle testing.The test of the forward vehicle in identification system and method is mainly discussed,including the relevant national standards referenced,the test system construction and test environment,the test plan design and the actual road environment test and subjective evaluation.The real vehicle road test verifies the designed hardware and software programs,and the algorithm has good stability and can effectively identify different vehicles on the road.The implementation of this project opens up new ideas for the development and application of embedded single-core ARM chips in ADAS-related products.This low-cost,cost-effective embedded software and hardware platform will bring positive effects to the large-scale application and development of FCW and AEB system technology in the future.