Robust Recognition of Truck License Plate in Mine Environment

To improve the decreased recognition performance for truck license plate in complex mine environment, which is caused by such factors as polluted, damaged, a robust license plate recognition method is proposed. Firstly, several candidate locations of license plate in the truck image are obtained by utilizing features of both edge and color. Furthermore, the parameter of license plate area intersection ratio is defined to find the precise localization among those above candidates. Then, a character segmentation scheme based on the gray projection method and morphological operator is used to remove those interference factors, such as frame, rivet and stain and uneven illumination. Finally, to increase the robustness of characters recognition by CNN, a self-built license plate character dataset is constructed, which contains various polluted samples obtained by manual simulation. As shown by the experimental results, the proposed method has higher accuracy in license plate location, and obtains an improvement on character recognition accuracy by 4% compared to other existed methods.