Saliency-based traffic sign detection method

Abstract

The invention discloses a traffic sign detection method based on saliency, and provides an end-to-end convolutional neural network model. According to the method, a traffic sign detection task is optimized, and a saliency detection model with a VGGNet-based feature pyramid network as a trunk is provided. The network extracts information of different hierarchies in pictures from feature maps of different scales in a bottom-to-top process, and fusion is carried out in top-to-bottom channels of the pyramid network. By adding the feature aggregation module before each fusion operation in the top-down path, seamless fusion of high-level semantic information in a deep-level feature map contained in the global guidance module and shallow-level features can be facilitated. Through the two modules established on the pooling operation, high-level semantic information can be refined step by step, so that the saliency map generated by the network model has richer details. The result is a grey-scale map with the pixel value being [0, 1], 1 in the map represents the area where the traffic sign is located, 0 represents the background area, and the task of detecting the traffic sign is successfully achieved.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a traffic sign detection method based on saliency. Background technique [0002] Road traffic signs are road facilities set up by the national transportation department on both sides of the road, using signs with text or special symbols to convey guidance, restrictions, warnings or instructions, and play a role in directing vehicles to move forward and conveying traffic regulations. Therefore, it is very important for drivers to accurately and quickly recognize the meaning of road traffic signs. With the high-speed economic development of our country, the number of all kinds of motor vehicles has increased rapidly, and the traffic problems caused by it have become more prominent. However, one of the ways to solve traffic problems is to develop intelligent transportation systems. From the perspective of national development strategies, intelligent transportation systems...

AI Technical Summary

Problems solved by technology

[0008] Different from the common deep learning-based target detection, most of these methods need to extract a large number of target candidate areas and send them to the classification network for judgment and recognition. The requirements for calculation are very high, and it is difficult to run in real time on general equipment.

In addition, the current traffic sign detection method based on deep learning requires a large data set of test pictures, and the problem of positioning small objects has not been solved.

Images