Traffic sign detection method based on small target feature enhancement

Abstract

The invention discloses a traffic sign detection method based on small target feature enhancement. The method mainly comprises the following steps: constructing a traffic sign data set and performing data enhancement; for the small size of a detection target in the data set, a prior frame is obtained by using a K-means + + clustering algorithm, and a nonlinear clustering distance is used; the network structure is optimized in a targeted mode according to the problem that many small targets appear in the data set; according to the problem of serious unbalance of positive and negative samples of data, a loss function of an algorithm is optimized in a targeted manner, and dynamic weighting of a target is realized. According to the method, traffic sign detection is realized in an urban streetscape scene, small target detection recall and precision improvement can be realized by improving a network structure, a loss function and the like of an algorithm, enhancing fine-grained features of a target and the like, and the method has relatively high accuracy on detection of traffic signs with many small sizes.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to a traffic sign detection method based on small target feature enhancement. Background technique [0002] With the rapid development of artificial intelligence technology, the intelligent driving industry is also developing rapidly, especially the development of deep learning has made vehicles achieve great success in perception and positioning. [0003] Traffic signs contain important road traffic information, which is an important part of vehicle environment perception. The detection accuracy of traffic signs is an important indicator for the actual deployment of algorithms to vehicles. [0004] Among the existing detection methods, they are mainly divided into two categories. The first type is a two-stage detection algorithm represented by Faster-Rcnn. This type of algorithm divides the entire detection process into two parts. The first step is to train the RPN networ...

AI Technical Summary

Problems solved by technology

Although the accuracy of the two-stage algorithm is higher than that of the one-stage algorithm, its speed cannot meet the real-time requirements. As the latest version of the Yolo series, Yolov5 has a good performance in detection speed and accuracy, but due to There are many small targets in the traffic sign dataset and the imbalance between positive and negative samples is serious, resulting in poor detection performance of Yolov5 on the traffic sign dataset TT100K

Images