Traffic sign detection and recognition method based on improved YOLOv3-tiny

Abstract

The invention provides a traffic sign detection and recognition method based on improved YOLOv3tiny. The method comprises the following steps: collecting traffic sign image data, carrying out the enhancement and amplification of the image data through geometric transformation and color transformation, and carrying out the image labeling, thereby obtaining a traffic sign training set; constructingan improved YOLOv3tiny network model, and training the improved YOLOv3tiny network model by adopting the traffic sign training set; and constructing a traffic sign test set according to the traffic sign image data, and detecting and recognizing the traffic sign test set by using the trained improved YOLOv3tiny network model. The improved YOLOv3tiny provided by the invention has relatively strong generalization capability, occupies relatively small storage space and video memory space, improves the detection and recognition accuracy, can also ensure the real-time performance, and can realize accurate and quick traffic sign detection and recognition in vehicle-mounted embedded equipment with limited computing power.

Description

technical field [0001] The invention relates to the technical field of deep learning of artificial intelligence, in particular to an improved traffic sign detection and recognition method based on YOLOv3-tiny. Background technique [0002] Deep learning object detection technology uses convolutional neural network for feature extraction, and through training and learning, it achieves more powerful adaptability and generalization ability. The YOLOv3 network model uses multiple scales for prediction. The prediction performance represents the current top level in the field of target detection. However, its network model is relatively large and cannot meet the real-time requirements in vehicle-mounted embedded devices with limited computing power. The streamlined version of YOLOv3-tiny, which focuses on detection speed, has high real-time performance and takes up less memory space, but has the problem of low detection accuracy and cannot be fully applied to traffic sign detectio...

AI Technical Summary

Problems solved by technology

The YOLOv3 network model uses multiple scales for prediction. The prediction performance represents the current top level in the field of target detection. However, its network model is relatively large and cannot meet the real-time requirements in vehicle-mounted embedded devices with limited computing power. The streamlined version YOLOv3-tiny, which focuses on detection speed, has high real-time performance and takes up less memory space, but has the problem of low detection accuracy and cannot be fully applied to traffic sign detection tasks for automatic driving.

Images