Traffic sign detection positioning method and system

Abstract

The invention provides a traffic sign detection positioning method and system, and the method comprises the steps: employing a target detection algorithm to process a to-be-detected traffic sign image, and obtaining a labeling file outputted by the target detection algorithm; cutting the to-be-detected traffic sign picture based on the annotation file to obtain a cut picture containing the trafficsign; performing color segmentation processing on the cut picture to obtain candidate regions corresponding to different colors, wherein the colors are determined by the colors of different types oftraffic signs; and determining a target positioning coordinate based on the candidate area, and writing the target positioning coordinate into the annotation file. The characteristic that the trafficsign has an obvious color characteristic is utilized; firstly, a picture to be detected is input into a target detection algorithm to obtain a picture containing a traffic sign, the region is cut out,and accurate positioning is carried out by using a color segmentation algorithm, so that the interference of objects with other colors in the original picture of the traffic sign to be detected is avoided, and the target positioning accuracy can be greatly improved compared with the traditional target detection algorithm.

Description

technical field [0001] The present invention relates to the field of image processing, and more specifically, to a traffic sign detection and positioning method and system. Background technique [0002] Since Ross Girshick proposed RCNN in 2014, the target detection algorithm has experienced rapid development for several years. Algorithms such as the RCNN series, SSD and YOLO have emerged one after another, making target detection tasks that were difficult to complete in the past easier. In academia or most industrial fields, classification performance is the main evaluation index of target detection algorithms, while positioning generally only needs to achieve IoU>60%. However, in some applications, positioning accuracy is a more important performance indicator, such as automatic annotation tools. [0003] Deep learning datasets often include tens of thousands of images, and image annotation is a very time-consuming task. The development of automatic annotation tools c...

AI Technical Summary

Problems solved by technology

The underlying algorithm of the automated labeling tool is also a target detection algorithm, but the tool has very high requirements on the algorithm's positioning ability. Generally, the acceptable positioning deviation is about 3 to 5 pixels. Otherwise, manual correction is required, and the purpose of improving efficiency cannot be achieved.

Images