Research On Typical Obstacle Detection And Recognition Methods Based On Vehicle Vision System

Detection and identification of typical traffic barriers on a road constitute an important part in the intelligent driver assistance system.However,due to some unfavorable factors involved in typical obstacles on traffic roads such as information mixing,occlusion omission and illumination influence,related technologies are still confronted with numerous unsolved difficulties.In light of the aforementioned disadvantages,this paper proposes a robust and reliable method for detection and identification.The whole system is divided into two sections:detection and identification.In the detection section points the following steps:1.Introduced first were the common detection methods of traffic barriers,including the detection method based on color space model and the detection method based on edge feature extraction.2.Then a detection method based on a combination of MSER and MSCR was proposed for complicated scenes.3.At the experimental stage,regions of interest were segmented by extracting color features and edge features,and the detection system entered the feedback stage when it was difficult to determine a region of interest.4.At the feedback stage,with a combination of MSER and MSCR,the final region of interest was determined through multiple feedbacks,hence a completion of the entire detection process.5.Additionally,during the experiment,detection results of different color spaces and algorithms used in different stages were compared so as to verify the reliability and accuracy of the entire detection system.In the section of traffic barrier identification points the following steps:1.gradient features of images were extracted to compose a histogram of gradient features.In order to obtain a good classification effect,a comparative experiment was performed on the methods of feature extraction.2.Then,after comparing some aggregative indicators such as the accuracy of classifier algorithm and the difficulty of modeling,SVM algorithm was selected as the classifier for identification.3.The experiment compared the overall identification results under different kernel functions and adjusted the training samples of relevant parameters so as to accommodate to the actual road conditions in our country.The results of the experiment in this paper have proved the feasibility and validity of the related algorithms.