Research On Environmental Perception Method Of Campus Patrol Car Based On Monocular Vision

With the development of science and economy in the world,more and more industries are developing towards the direction of intelligence and automation.As an emerging direction in the 21 st century,machine vision has been widely used in all walks of life,which greatly saves labor and work costs.In addition to the application in medical treatment,agriculture,industry and other aspects,with the rise of intelligent driving,the intelligent vehicle environment awareness method based on machine vision has become a hot research topic today.In the intelligent vehicle environment awareness system,how to accurately identify the road condition information and locate the detected object is the primary condition to realize the automatic obstacle avoidance of the intelligent vehicle.Both target detection and ranging technology are important links of intelligent transportation and vehicles,and have a wide range of application scenarios.In this paper,targets such as vehicles and pedestrians are detected and located in real time,so that patrol vehicles can get familiar with the surrounding environment information and make timely obstacle avoidance planning.In order to improve the adaptability of patrol vehicles in different environments,this paper mainly improves the traditional feature detection method so as to improve the detection accuracy and efficiency.The main work includes the following four aspects:(1)Pedestrian detection based on CENTRIST feature.Compared with the HOG feature detection,the CENTRIST feature greatly saves the computation time and storage space.Based on the traditional SVM classifier,a cascade classifier composed of S VM linear classifier and HIK SVM was proposed to improve the detection accuracy.A fast detection algorithm based on these two classifiers is studied.The time complexity of the classifier is reduced by using auxiliary images and tables.(2)An improved lane line detection algorithm based on Hough feature is implemented.Aiming at the detection of possible lane line area,a detection method based on the brightness difference between lane line and both ends is proposed.Compared with Canny edge detection and OUST,the brightness difference method used in this paper has less calculation cost,faster speed and more complete information retention.In addition,conditional screening is carried out on the detected lane lines according to the campus environment to filter out the unnecessary lane line information.(3)Vehicle detection is realized based on Adaboost cascade classifier.Based on vehicle contour and symmetry,the Adaboost weak classifier was trained by using Haar feature.After several iterations,the Adaboost strong classifier was obtained.The effectiveness of the classifier was verified by vehicle detection results under different environments.(4)single-eye distance measurement method was designed and verified.According to the imaging principle and coordinate transformation model,the distance measurement method based on the width of target detection object and the method based on the contact point between target object and ground are proposed.The advantages and disadvantages of the two methods were evaluated through experiments,and the overall experimental results show that the measurement accuracy of the two methods based on the campus environment meets the requirements of detection.