Traffic sign recognition method based on capsule neural network

Abstract

The invention relates to a traffic sign recognition method based on a capsule neural network. The method comprises the following steps: preprocessing an image by adopting methods such as image equalization, maximum stable extremum region segmentation, normalization and the like, eliminating interference of factors such as motion blur, background interference, illumination, local occlusion damage of a traffic sign and the like, and segmenting an image of a region of interest, so that the image of the region of interest can be effectively extracted, the recall ratio of a weak light condition isimproved, and the robustness is enhanced; in addition, a capsule neural network structure is introduced, convolution layer bottom layer features are adopted, a vectorized capsule unit is packaged after passing through a main capsule layer tensor vector, weight parameters are updated through dynamic routing clustering and back propagation, model training and model weight parameter outputting are achieved, the training speed is high, and the training time is shortened; and finally, image classification is realized according to the trained model weight parameters and dynamic routing clustering, so that the recall ratio of weak light pictures can be effectively improved, and the recognition rate of traffic signs is improved.

Description

Technical field: [0001] The invention relates to the technical field of traffic sign detection and recognition, in particular to a traffic sign recognition method based on a capsule neural network. Background technique: [0002] With the development of social economy, automobiles have become the most used means of transportation in the world, resulting in frequent traffic accidents and increasingly serious traffic jams, resulting in a large amount of economic losses. Faced with this problem, automakers, academia and government experts have worked together to develop advanced intelligent transportation systems to improve and strengthen traffic safety. Therefore, intelligent transportation systems have developed rapidly, and road traffic sign recognition is a difficult task in the field of intelligent transportation. One of the bigger problems. In today's in-vehicle systems, the prompts of traffic signs are mostly obtained through digital map data, but this method is limited ...

AI Technical Summary

Problems solved by technology

The above methods all use CNN neural network to train and recognize traffic sign images, but using continuous pooling layers will lose some valuable information such as space, making it difficult to recognize the same image after changes such as rotation, flipping, and translation

[0005] To sum up, the current machine learning methods focus on manually extracting image features, and feature extraction is more complicated and requires a lot of manpower and time.

Neural network methods such as CNN will lose part of the information of the image, resulting in the need to consider rotation, flip, translation and other changes in the construction of the training data set, which increases the amount of calculation of the method

Images