Study On The Mark Recognition Algorithm Under Variable Illumination For CyberCar System

CyberCar is a kind of low-speed electricity-powered intelligent vehicle, Somesuch vehicles can be administered and deployed by a control center. The vehicle isfree of noise and pollution with the characteristic of high safety and convenience. Ithighly represents the human-oriented ideology in transportation system. Withhigher demand of transportation safety and environmental protection and with theimprovement of social information and automation, CyberCar will be put intooperation in more fields. It will eventually develop into an important symbol withwhich social civilization and technical advancement is measured.Image processing and pattern recognition is the key technology to CyberCarbased on computer-vision. This paper is dealing with outdoor computer-visionbased CyberCar in commercial use. The research work in this paper is mainlyrelated to such technology as image processing and pattern recognition forcomputer-vision based CyberCar.Computer Vision is featured by abundant information and high intelligence. Inrecent years, it has been widely used in self-guiding vehicles. JLU IV group hasbeen developing a new type of self-guiding vision-based vehicle which usesspecially designed white strap-shaped paths and different marks. The path is tracedby the vehicle to navigate and steer .The marks of different kinds are recognized tocontrol the vehicle speed automatically. In this process, ACCD camera mounted onfront of the vehicle is to get image information forward. By computing andanalyzing continuous image frame information, the built-in computer can deteminethe actual condition forward in order to realize stable path-tracing andspeed-controlling.The title of this paper is "Study on the Mark Recognition AlgorithmUnder Variable Illumination for CyberCar System", the paper includes five parts:1:study on the image pre-processing algorithm; 2:Study on the fusing andanalyzing of automatic image segmentation algorithms;3: Speed-control marksrecognition for CyberCar;4: Number marks recognition;5: Experiments onalgorithm of marks recognition;Several automatic image segmentation algorithms for variable illuminatingconditions are discussed in the paper. Based on comprehensive analysis of thealgorithms, the paper in particular presents both a real-time improved and asegmenting performance improved algorithm. Original OTSU algorithm performspoorly in segmenting a frame image in term of real-time. In addition, the finalsegmentation result of higher illuminating image is poor. Time improved algorithmby using different evaluation function dramatically reduced the time insegmentation. Meanwhile, Segmenting performance improved algorithmsuccessfully solves the problem of image segmentation under higher illumination. In view of the failure of several thresholding algorithms in segmenting imageunder intensive illumation, the paper presents a new algorithm based on Robertregional operator. This algorithm makes use of the special regional features of theimage instead of a threshold to segment the image. A new algorithm is proposed based on grayscale histogram. Aspecial filteringmethod is developed according to special histogram distribution feature.Byfiltering before segmentation for images under complex road background, thealgorithm is to determine a threshold for the image based on dual-hump-shapedhistogram. Several popular image segmentation algorithms are programmed toexperiment and analyze their performances. Image grayscale mean value μT is taken as mathematical criterion todescribe the actual illuminating condition. All the algorithms are analyzed in detailwith this criterion. Another criterionθis proposed to evaluate the performances ofdifferent algorithms. 25 typical illuminating conditions are included to get morereasonable evaluations for all the algorithms. By comparing both the performances and real-time of all algorithms, a fusingcompound algorithm is obtained to solve the problem of image segmentationwithin the mean value of from 10 to 250. A matching recognition algorithm based on four features:standardized markarea , relative vertical coordinate to center of gravity, vertical and horizontal offsetis raised. Due to image distortion of the mark, the recognition algorithm is speciallyconsidered to divide the whole image into upper, lower and middle sub-regions.The negative effect of distortion is fairly reduced.