A Method of Infrared Road Scene Segmentation Based on Class Prototype Regression

Abstract

The invention relates to a method for segmenting infrared road scenes based on category prototype regression, comprising the following steps: 1. Category prototype feature regression: using a large number of data labels and depth features to obtain category feature prototypes through regression; 2. Constructing a relationship matrix: obtaining After obtaining the category feature prototype, construct a relationship matrix through the depth feature and category feature prototype; 3. Attention enhancement: construct different attention maps through the relationship matrix to achieve feature enhancement; 4. Build attention modules: establish category attention modules and space Attention module, which aggregates the functions of two attention modules. The present invention proposes a category prototype regression strategy to regress the entire data set to obtain representative category prototype features, and at the same time cluster network depth features to make the global category features more compact; at the same time, the differences between categories are enlarged, and corresponding construction The relationship matrix and attention module make the overall features more compact and improve the final image segmentation accuracy.

Description

technical field [0001] The invention relates to an infrared road scene segmentation method based on category prototype regression, and belongs to the technical field of image processing. Background technique [0002] Compared with other scenes, vehicle road scenes are more complex, and many problems may occur in complex scenes, such as complex backgrounds that make target recognition more difficult, or similarities between targets that interfere with visual features , Different targets are misclassified, especially in the infrared vehicle road scene, the edge of the target is weak, and the boundary line between the background and the foreground is not obvious, which will lead to the accuracy of visual features. Therefore, in order to achieve higher recognition accuracy, the segmentation model needs to have stronger discriminative ability for weak edges and similar objects. At present, image semantic segmentation technology is mainly aimed at classification tasks at the pixe...

AI Technical Summary

Problems solved by technology

[0004] At present, the most complex vehicle-mounted road scene is the urban street scene, because in the street, the scene is complex and changeable, the background and the target are mixed together, it is difficult to distinguish, and the traffic volume on the road is large, sometimes there is dense situation, the general algorithm is difficult To achieve higher precision, at present, it is difficult for intelligent driving to achieve zero errors

Images