Road surface traffic sign recognition method based on convolution neural network

Abstract

The invention discloses a road surface traffic sign recognition method based on a convolution neural network, and the method comprises the following steps: image collection and preprocessing; and convolution neural network structure design and training. The method employs a V-parallax method to obtain a road surface area from an original image, can reduce the impact caused by non-road-surface interference, and improves the extraction precision of a road surface area. The invention employs a vertical view to reconstruct the road surface area, enables unparallel lines, presented in a visual image because of a view angle, to be reconstructed into approximately parallel lines, facilitates the recognition of a road surface traffic sign, and improves the capability of adapting to view angle inclination. The method the deep learning method (convolution neural network), can extract recessive characters reflecting the data essence from a large number of training samples. Compared with a shallow-layer learning classifier, the method is higher in learning efficiency and recognition precision.

Description

technical field [0001] The invention belongs to the field of automobile safety assisted driving, and in particular relates to a method for recognizing road traffic signs. Background technique [0002] Road traffic sign recognition is an important research content in the research of automobile safety assisted driving system, and it is an effective way to ensure safe driving and reduce traffic accidents. [0003] At present, the methods of road traffic sign recognition can be roughly divided into shape-based recognition methods, color-based recognition methods and machine learning-based recognition methods. The shape-based recognition method has poor adaptability to viewing angle tilt. Color-based recognition methods cannot achieve satisfactory results in environments with a lot of external interference. The recognition method based on machine learning mainly uses a classifier to classify the target, generally there are shallow learning models such as artificial neural netwo...

AI Technical Summary

Problems solved by technology

However, these methods need to extract the explicit features of the target, which is likely to cause the loss of important information and poor environmental adaptability; moreover, the limited learning depth makes it difficult to ensure the efficiency of the learning structure in the case of huge parameters and samples

Images