Task-driven traffic sign saliency detection method in natural scene

Abstract

The invention relates to the field of computer vision, and discloses a task-driven traffic sign saliency detection method in a natural scene, and the method comprises the following steps: S1, collecting training data; s2, inputting images in the training set data, extracting total global features of the images by using a convolutional neural network, and extracting feature information of the images under a plurality of different resolutions; performing multi-layer extended convolution learning on the feature information of the image under different resolutions by using an extended convolutionnetwork to extract features and contrast features; s3, performing up-sampling learning on the features and the contrast features to obtain a feature map under each resolution, and then fusing the feature maps into a head office feature; s4, performing prediction to finally obtain a traffic sign saliency feature map; s5, repeating the step S2-S4 to train the convolutional neural network, and storing a training model; s6, inputting a to-be-predicted image, and obtaining a traffic sign saliency feature map of the to-be-predicted image. The traffic sign detection precision is improved, and the method can be widely applied to the field of unmanned driving.

Description

technical field [0001] The present invention relates to the field of computer vision, in particular to the technical field of image saliency detection based on deep learning, and more specifically, to a method for saliency detection of traffic signs in natural scenes based on task-driven. Background technique [0002] With the increase in the number of cars, traffic problems are becoming more and more serious. Recognition of traffic signs is the most important problem in the driving process, and it is also very important for road maintenance, driver assistance systems and self-driving cars. [0003] For example, many practical factors need to be considered in the development of advanced driver assistance systems (ADAS), the most basic of which is traffic sign recognition (Traffic Sign Recognition, TSR). TSR is a difficult real-scene graphics recognition problem. Its main function is to provide road information to the driver and remind the driver to make reasonable operations...

AI Technical Summary

Problems solved by technology

However, in real-world scenarios, when specific targets need to be detected, for example, applications such as automatic driving need to detect specific targets, such as traffic signs that drivers need to pay attention to. Using the first method for detection, it is very likely that only the most The obvious traffic sign, but it is impossible to display all the traffic signs in the image. Therefore, it is necessary to provide a task-driven traffic sign detection method in natural scenes to achieve accurate detection of traffic signs.

Images