Research On The Light Interference Suppression Technology Of Vehicle Target Detection In Tunnel Scenes

Vehicle targets detection is the foundation and premise of video detection of traffic events, such as stops and congestion. In tunnel scenes, because of the ambient light and vehicle light, the extraction of the vehicle targets will be affect, greatly influenced the accuracy of vehicle target extraction. While existing methods have shortcoming that is difficult to optimization. Therefore, to solve the ambient and vehicle illumination interference in tunnel scenes, the research into the suppression of illumination interference in vehicle target identification has great significance theoretically and practically.Based on the analysis on the problems and difficulties in light interference suppression, the thesis mainly researches the background modeling in tunnel scenes, ambient light interference suppression and vehicle light interference suppression. A novel of light interference suppression technology in tunnel scenes based on video is established.Aiming at the existing vehicle target problem of the interference of background modeling, this paper provides a method of the background modeling based on the characteristics of the pixel neighborhood frame. In order to solve this, considering the reliability of background models, it is natural to choose the motion features of pixel point neighbors as the judgment whether they can be seen as pixel points, and transform them as background modeling weighted value to improve traditional background modeling methods. Consequently, the better extraction of vehicle targets is realized.Aiming at the false vehicle target area problem caused by tunnel ambient light shifting, this paper provides a method of the suppression of illumination interference based on the area discreteness. A illumination interference distance-light intensity model is made by analyzing ambient light shifting features; utilizing the model to process scatter points fit for the foreground pixel value after motion division, this method chooses coefficient of variation as discrete index to assess the discrete degree in different areas; combining probability distribution relationships of coefficient of variation to select illumination area, this method eliminates the illumination interference in vehicle target foreground, achieving the suppression of ambient light interference. The results show this method suppresses the interference of ambient illumination efficiently, and decreases the false detection rate of vehicle target identification.To solve the problem of low accuracy of extraction of vehicle targets in vehicle light illuminated areas in tunnel scenes, this paper proposes a method based on gradient features, and builds a light intensity model of vehicle light illumination on roads according to the optical radiation and space structure features, constructing a vehicle light gradient function. On basis of the invariance of its gradient directions, a foreground mask is made by analyzing the screened non-vehicle light illumination area. The research results show this method can eliminate light interference sufficiently, and improve the accuracy in extraction of vehicle targets.In the end, to summarize the research achievements above-mentioned, this paper proposes a set of vehicle target identification light interference suppression technique in tunnel scenes which can eliminate the light interference in tunnel scenes. This algorithm is universal, and has great theoretical and practical significance. And applied to the parking incident detection system, the application effect shows that the proposed light interference suppression technique can reduce parking error detection rate of light interference, and has good adaptability.