Fast R-CNN-based pavement traffic sign identification method

Abstract

The invention discloses a Fast R-CNN-based pavement traffic sign identification method. The method comprises the following steps of: carrying out image acquisition and preprocessing, and making a sample set; inputting a training set and carrying out multi-task training on a Fast R-CNN network; enabling a to-be-identified picture to pass through a plurality of convolutional layers and pooling layers so as to obtain a feature map; obtaining a corresponding feature frame from candidate frames, and respectively obtaining two output vectors: a classification score and window regression, through an ROI pooling layer and a total connection layer; and carrying out non-maxima suppression processing on all the results to generate a final target detection and identification result, so as to identify traffic signs. According to the method, a Fast R-CNN deep learning method is adopted, so that a redundant feature extraction operation in a regional convolutional neural network R-CNN is avoided, multi-task training is realized, an extra feature storage space is not required, and the detection speed and precision are improved. Compared with shallow-layer learning classifiers, the method has higher learning efficiency and identification precision.

Description

technical field [0001] The invention belongs to the field of image processing and automobile safety assisted driving, and in particular relates to a method for detecting and recognizing road traffic signs based on Fast R-CNN, which is used to solve the problem of low recognition accuracy in road traffic sign recognition. Background technique [0002] Traffic Signs Recognition (TSR, Traffic Signs Recognition), as an important branch of vehicle assistance systems, is one of the unsolved problems. Since traffic signs contain many important traffic information, such as the current driving speed prompt, the change of the road ahead, and the driver's behavior constraints, how to quickly, accurately and effectively identify the traffic on the road in this auxiliary system Signs and feedback to the driver or control system have very important research significance for ensuring driving safety and avoiding traffic accidents. [0003] Commonly used methods for road traffic sign recogn...

AI Technical Summary

Problems solved by technology

Shape-based recognition methods are less robust and perform poorly in complex environments

The method of combining feature extraction and classifier has better recognition effect, but the calculation cost is large and the environmental adaptability is relatively poor.

Images