Method and system for recognizing vehicle number by using deep learning image recognition-based vehicle detection and tracking technology

摘要

A vehicle number recognition method using a vehicle detection and tracking technology based on deep learning image recognition is disclosed. The vehicle number recognition method using the deep learning image recognition-based vehicle detection and tracking technology is a step of generating a plurality of image frames by a camera included in a vehicle license plate recognizer successively shooting a shooting area, the license plate recognizer is deep learning Detecting a vehicle in the plurality of image frames using an algorithm and generating a bounding box including the vehicle, wherein the license plate recognizer is t( t is a natural number greater than or equal to 4) performing an estimation operation of estimating the position and size of the bounding box in the (t+1)-th image frame using image frames including up to the (t+1)-th image frame, the license plate recognizer is the performing a similarity calculation operation of calculating a similarity between the bounding box generated in the (t+1)th image frame and the estimated bounding box, wherein the license plate recognizer determines the (t+1)th image frame according to the similarity performing a matching operation of matching the generated bounding box with the estimated bounding box with each other, the license plate recognizer determining whether a vehicle included in the matched bounding box arrives at an arbitrary location, and the matched bounding box When it is determined that the vehicle included in the bounding box arrives at an arbitrary position, the license plate recognizer performs a license plate region extraction operation of extracting the license plate region of the vehicle from the (t+1)-th image frame, and the extraction Recognizing the vehicle number in the license plate area, after the bounding box including the vehicle is generated, the license plate recognizer counts the number of bounding boxes generated in each of some image frames among the plurality of image frames When two or more bounding boxes are counted in each of the image frames of some of the plurality of image frames, the vehicle number recognition device performs the estimation operation, the similarity calculation operation, and the matching operation , some image frames among the plurality of image frames When only one bounding box is counted in each, the license plate recognition device determines the size of the bounding box in the first image frame among the some image frames and the size of the bounding box in the last image frame among the some image frames. Comparing the size of the bounding box, when the value of the comparison result is less than a certain threshold value, the license plate recognition device omits the estimation operation, the similarity calculation operation, and the matching operation, and the license plate of the vehicle performing a region extraction operation, and when the value of the comparison result is greater than the arbitrary threshold value, the vehicle number recognition apparatus performs the estimation operation, the similarity calculation operation, and the matching operation, and performing a license plate area extraction operation.