| In recent years,intelligent vehicles have developed rapidly on account of their characteristics of electrification,networking,driverlessness,and lead the future of smart drive.Intelligent driving is the core function of intelligent vehicles,which plays a key role in enhancing the competitiveness of the automobile industry,accelerating the construction of an intelligent transportation system,innovating car models,improving driving safety,etc.Nevertheless,the environment awareness module based on the vehicle sensor system is the prerequisite forintelligent driving,which has effects on the reliability and accuracy of follow-up plans and control tasks.Consequently,it is of crucial significance to study the environmental perception technology of intelligent driving.This paper takes the environmental perception technology of intelligent vehicles as the research objec,carrying out subtasks by algorithm design such as parking space detection,lane detection and vehicle detection,and discuss algorithm performance in actual scene.The main research contents are as follows:First,the research puts forward a practical improved parking space detection algorithm so as to solve the problem of parking space information input in automatic parking system.A gain compensation for ultrasonic radar amplitude attenuation is designed,and the fitting process of characteristic line segments based on ultrasonic data is introduced.The defects of the traditional parking space detection algorithm are elaborated in detail,and a parking space detection algorithm based on fitted characteristic line segments is proposed.The experimental results show that the average detection accuracy of the parking space detection algorithm proposed in this paper reaches 92.5%,which verifies the reliability and accuracy of the algorithm.Second,the lane detection algorithm for structures roads is studied,the basic theories related to image preprocessing are introduced,and the Hough line detection algorithm and line segment detector LSD are compared and analyzed.The rough contour line segments detected by LSD are post-processed,including filtering,merging and other operations,searching for lane candidate fitting points on the improved binary image and combining the RANSAC algorithm to fit the lane model.The experimental results show that the proposed lane detection algorithm can achieve ideal performance in both detection rate and robustness.Finally,the vehicle detection based on YOLO v3 is studied,the data set is constructed by manual annotation,and the k-means algorithm is used to cluster the target border of the data set,so as to design a more reasonable candidate border dimension.The optimized YOLO v3 convolutional neural network is used to train the recognition models of different types of vehicles such as cars,motorcycles,buses and trucks.Experiments show that the performance of the optimized YOLO v3 algorithm in vehicle detection can meet the needs of vehicle applications. |
