R-CNN-based license plate locating method

Abstract

The invention belongs to the technical field of license plate locating, and discloses an R-CNN-based license plate locating method. The R-CNN-based license plate locating method comprises the steps of extracting license plate features through a CNN, and selecting a foreground candidate box by utilizing the extracted license plate features through an RPN; accurately correcting a license plate position through a target classifier and a coordinate regression device; and through training of a large amount of data, obtaining a license plate locating model with relatively high locating rate in a complex environment. An R-CNN license plate detection technology adopted in the method has very strong robustness and can detect license plates with blue, yellow and white backgrounds and different specifications in environments of foggy day, rainy day, light reflection, night, license plate contamination and damage, license plate inclination and the like. A detection model adopts offline training and online testing, so that the detection speed is more efficient than that of a conventional license plate locating algorithm.

Description

technical field [0001] The invention belongs to the technical field of license plate location, in particular to an R-CNN-based license plate location method. Background technique [0002] At present, in the research of license plate location, most algorithms of license plate location are based on the traditional Hough transform, color features or shape features. These design features are considered to have high positioning accuracy for a specific environment, but they are not very stable in complex actual environment applications. For example, using the Hough transform algorithm to detect the rectangular frame of the license plate, it is difficult to complete the license plate location when the license plate is damaged or the frame is blurred; the license plate location based on color features encounters strong reflections or the color of the license plate is similar to the license plate. Usually it fails; in the license plate location method based on morphology, the licens...

AI Technical Summary

Problems solved by technology

[0003] To sum up, the problems existing in the existing technology are: the traditional license plate location methods all use artificially set features, which are relatively single, and must be adapted to complex environments such as foggy days, nights, license plate stains, and license plate tilts at a large angle. , the traditional algorithm is still unable to satisfy

Images