Small-sized traffic sign recognition method based on yolov3 and asymmetric convolution

Abstract

The invention relates to a small-sized traffic sign recognition method based on YOLOV3 and asymmetric convolution, comprising the following steps: preparing a data set containing different types of traffic signs and performing data enhancement; building a network and training: building and adding asymmetric convolution YOLOV3 improved network of module and spatial pooling pyramid module; save the parameters of the improved network as a model, which is the model of the unfused asymmetric convolution module; the asymmetric convolution module in the fusion model: read the unfused asymmetric The parameters saved in the model of the convolution module are calculated to fuse the three parallel 3×3, 3×1 and 1×3 convolution kernels of the asymmetric convolution module in the module into a 3×3 convolution kernel; Four steps, detection and recognition of traffic signs.

Description

technical field [0001] The invention relates to the technical field of intelligent driving, in particular to the field of traffic sign detection and recognition. Background technique [0002] With the continuous development of the economy, the appearance of automobiles has greatly increased the convenience of our travel, but at the same time it has also brought about frequent traffic accidents. The main causes of traffic accidents are: illegal driving, fatigue driving, substandard road construction and so on. In order to increase safety and reduce the occurrence of traffic accidents, people have made a lot of efforts. Early car safety tended to protect the occupants of the car after a vehicle collision, while today's cars are more inclined to prevent accidents, so Advanced Driver Assistance Systems (ADAS) have gradually developed, in which the detection of traffic signs And identification is a very important content in ADAS. Traffic signs transmit guidance, restriction, w...

AI Technical Summary

Problems solved by technology

[0004] There are still the following problems in the existing methods: traditional methods and machine learning-based methods are less robust and difficult to apply in practice; deep learning-based methods are more robust, but their accuracy and speed are difficult to balance. The method with high precision is complex and has a large amount of calculation, while the method with fast calculation speed has low precision

In the existing public data sets GTSDB and GTSRB, traffic signs are only divided into four categories, which obviously cannot meet the needs of actual intelligent driving; secondly, traffic signs account for a large proportion in each image, which is very important for detecting small images under high-resolution images. For the problem of size traffic signs, this dataset is not suitable for

Images