License plate detection system in natural scene based on deep learning

Abstract

The invention relates to a license plate detection system. A license plate detection system in a natural scene based on deep learning comprises an input unit used for inputting natural scene images, including, but not limited to, camera acquisition image input and code stream decoding input; a first deep network unit which is used for preliminary screening of license plates and comprises six convolution layers, a maximum pooling layer and a non-maximum suppression processing unit; a second deep network unit which is used for screening the license plates again and comprises three convolution layers, three full connection layers and two maximum pooling layers; an output unit which is used for comparing all the license plate areas which pass through the first deep network structure preliminary screening and the second deep network structure screening, and if overlapping exists and the overlapping area is larger than a threshold value Th _ 4, the areas are combined into one area; and the combined regions are output, wherein the output regions are suspected license plate regions. The method is small in occupied resource and suitable for being used by embedded equipment.

Description

technical field [0001] The invention relates to license plate detection, in particular to a license plate detection in natural scenes. Background technique [0002] To detect license plates in natural scenes, the traditional method is to filter by features such as color and edge, or to construct artificial features such as HOG and LBP for detection. The problem is that the detection is not accurate enough, and the false positive and false negative rates are high. After the rise of deep learning, people can also use deep learning frameworks such as SSD / YOLO / Faster RCNN to detect license plates, and license plate detection methods based on deep learning, such as 201610312822.2, 201710187201.0, 201710187289.6 and 201710531085.X. [0003] The problem with these methods is that the amount of calculation is large, and the CPU and DDR resources consumed are very large. However, due to the cost and earlier reasons of the detection equipment in natural scenes, the configuration will ...

AI Technical Summary

Problems solved by technology

[0002] To detect license plates in natural scenes, the traditional method is to filter through features such as color and edge, or to construct artificial features such as HOG and LBP for detection. The existing problem is that the detection is not accurate enough, and the false positive and false negative rates are high.

[0003] The problem with these methods is that the amount of calculation is large, and the CPU and DDR resources consumed are very large. However, due to the cost and earlier reasons of the detection equipment in natural scenes, the configuration will not be too high and the resources are limited. The above methods Not conducive to implementing applications on embedded devices with limited resources

Images