A traffic light recognition method based on tensorflow combined with multi-layer CNN network

Abstract

A traffic light recognition method based on TensorFlow combined with a multi-layer CNN network belongs to the field of computer vision and machine learning; prepare traffic signal recognition public dataset images, labels and video datasets; use OpenCV to recognize images in TensorFlow and traffic signal public datasets Change the size and output the normalized value of the three channels of RGB; transform and normalize the coordinate label according to the transformation rules of the graphics; extract the picture containing the traffic light information according to the label index and make it correspond to the label one by one Send pictures and labels into the CNN network to train and save the model; make the present invention effectively solve the technical problems that the accuracy of labeling is not high and the recognition speed is slow.

Description

Technical field [0001] The present invention belongs to the field of computer vision and machine learning, particularly to a traffic light based on the identification TensorFlow multilayer CNN in combination. Background technique [0002] The traffic light image location identification is a combination of computer vision and machine learning (MachineLearning) technology needs with the advent of artificial intelligence to automatically drive a car-based traffic information collection device for processing increasingly the larger, more efficient and how to find accurate traffic information within the range of the camera shown, relying solely on the traditional image recognition algorithm is efficient and accurate enough. For autonomous vehicles, in order to ensure the safety of driving under the existing laws, regulations and rules of the road, real-time awareness on the road is very important, one important point is to gather information processing power traffic lights, traffic li...

AI Technical Summary

Problems solved by technology

[0004] The present invention overcomes the above-mentioned deficiencies in the prior art, and provides a traffic light recognition method based on TensorFlow in conjunction with a multi-layer CNN network. The present invention uses traffic signal recognition public datasets, that is, Traffic Lights Recognition (TLR) public benchmarks as the most dataset, containing 6228 effective pictures and corresponding one-to-many labels make the image features extracted by the present invention have a certain expressive ability, and mark the traffic light positions in the pictures through the convolutional neural network algorithm, so that the present invention effectively solves the problem of The technical problems of low labeling accuracy and slow recognition speed

Images