The invention provides an automatic
license plate recognition method in a natural scene based on a
deep learning technology. The basic principle is as follows: firstly, using a lightweight MobileNet neural network as a
feature extraction network, adding the
feature extraction network to a
deep learning object detection algorithm SSD (
Signal Shot Multi-boxes
Detector), and training on a scene
license plate image; then, detecting a
license plate region by utilizing SSD-MobileNet and classifying the types of license plates; secondly, for the detected license plate area, determining a boundary
point set by searching for character contours through multi-threshold binarization operation, performing
line fitting on the boundary
point set to determine license plate corner points, and correcting the license plate in one step through
perspective transformation operation; and finally, sending the license plate to a
convolutional neural network with seven outputs to obtain all license plate character outputs. Compared with an existing license plate
recognition algorithm, the license plate detection and positioning
correction algorithm based on the
deep learning technology is faster and more accurate, and the method has robustness on license plate detection under complex natural scene. The
system does not need character segmentation operation, feature
transmission loss is reduced, the recognition speed is greatly increased through the end-to-end recognition network under the condition that the accuracy is guaranteed, the whole
system can achieve the real-time license plate recognition effect, and the practical value is achieved.