| Moving object detection and location is a focus in the field of image processing in recent years. The key technology is the feature extraction and description. The thesis summarizes the gray feature, texture feature and geometry feature, and chiefly analyzes the Maximally Stable Extremal Region (MSER) detection method. The MSER is affine-invariant to scale transformation, rotation transformation and transformation of the view-point. These affine invariances are verified by the experiments.The license plate character regions are the typical MSERs. Therefore license plates with different sizes, view-points and angles can be effectively detected. And then,a MSER license plate detection method is presented based on the priori knowledge of the license plate. Firstly the input image is pre-processed with grayscaling and gray stretching, etc; And then, the candidate MSER license plate regions are chosen by the pixel sum of the license plate character region; After that, the regions are processed according to morphologic and chosen by the scale and length-width ratio of license plate character; And then, based on the gray level jump and horizontal projection of license plate character, the single character regions are removed and the lower and upper borders of the license plate are determined in this step; At last, the left and right borders of license plate are determined by vertical projection. Based on MFC and OpenCV library, the effectiveness of the MSER license plate detection method is verified. The experimental results show that adopting the same threshold parameters, 450 images with size 720*540 with complex background in image library are tested, the license plate detection rate is 83.3%, the detection time of one image is about 400ms; In the vehicle face library extracted from one video sequence, 159 vehicle face images with size 322*131 are tested, the license plate detection rate is 95.6%, the detection time of one vehicle face image is about 40ms. |
PDF Full Text Request