Research On Shadow Detection Techniques For Moving Vehicles In Video Sequences

Video-based moving vehicle detection, location, classification, recognition and tracking have a broad application future in the Information Transportation System (ITS), while shadow detection is a very important part in the recognition and tracking of the moving vehicles.In the vehicle location systems, background subtraction and proper de-noising treatment is the usual way adopted to implement the segmentation of the vehicle. However, due to the existence of the shadows, neither of the above operations can separate the shadows from the vehicle object, so the vehicle obtained is the sum of the real vehicle and the shadows. In a word, the shadows can modify the shape of the object, thus introducing a distortion in the object detection process. While this will cause great inaccuracy even error in the following process such as the precise positioning, the extraction of the key part and the classification of the vehicles. From this point, shadow detection is a very important part in the recognition and tracking of the moving vehicles.By analyzing the usual algorithms for shadow detection, a new fast shadow detection approach for vehicles based on the moving region is presented in this paper. The moving region is obtained as follows: first using the background subtraction technique to extract the initial background model and determine the region that need to be updated according to the determination of the self-adaptive threshold, finally the moving region can be obtained by the subtraction between the image of the image sequence and the background. To wipe off the noises of the binary image of the moving region, a new fast de-noising method based on label is also presented in this paper. In the following shadow detection operation, a new method combing the color-based and model-based shadow detection methods is put forward in this paper. First establish the coarse model of the shadows, then extract the coarse region of the shadows combing the coarse model and moving region obtained through the above image pre-treatment, then impose the optimized shadow detection algorithm based on the HSV color space on the coarse region of the image only.The algorithm for shadow detection proposed in this paper can reduce the calculation amount substantially with the detection accuracy assured. Proved right and quite applicable by the experiments carried out in traffic pictures with moving objects in them, the algorithm is not sensitive to the changes in lighting and weather conditions, achieving our expectation.