Research On The Moving Object Detection And Tracking In Integrated Vehicle Image Sequence

Based on the computer vision, this paper constructs a vehicle drive auxiliarysecurity system to avoid the man-caused traffic accident and improve thesecurity of highway transportation. The key to this system is the detection andtracking of the lane and the vehicle, which are mainly researched in this paper. Taking into account the character of the highway and the perspective image,this paper adopts a straight model to describe the lane mark. Based on the whitemark, lane detection is performed by employing the color image segmentation,edge extraction and Hough transform. Based on the three features of a lane mark–starting position, direction andits gray-level intensity, a three-feature based lane tracking algorithm is proposed.The strength of the algorithm comes from fusing together three of the mostimportant features. Furthermore, an intelligent tracking strategy is embeddedinto the lane tracking algorithm, thus making the system more suitable to thecomplicated road conditions, for example, the lane mark is sheltered partly orfully. The vehicle detection algorithm takes the following two steps: hypothesisgeneration and hypothesis verification. At the first step, shadows under thevehicles are detected to estimate the possible presence of vehicles. At the secondstep, symmetry detection approach using intensity is exploited to verify theexistence of vehicles. This paper tracks the vehicles through image sequences by the repetition ofthe "matching-correction-prediction" strategy. When matching the template tothe edge image, the distance transform is used to overcome the drawback whichwould arise in traditional template matching methods. The distance transformcan allows more variability between the template and the edge image, thusmaking the algorithm more reliable. It can also enhance the matching efficiencysince it greatly reduces the calculation. Updating the template dynamically canmake the template more characteristic for the tracked vehicle and thereforeincrease the accuracy of the tracking. Moreover, Kalman filtering is employed topredict the position and size of the ROI in the next frame. It reduces the searchregion of the matching and decreases the chance of mismatches leading to amore rapid and robust tracking. Experimental results prove that the moving object detection and trackingalgorithm in this paper is robust, efficient and suitable for real-time processing.