Traffic road image segmentation method and system based on edge detection, and computer readable storage medium

Abstract

The invention provides a traffic road image segmentation method based on edge detection. The method comprises the following steps: inputting a traffic road image into a trained edge detection model; performing forward edge detection on the traffic road image to obtain an initial detection image; performing a plurality of convolution processes on the traffic road image through a plurality of convolution layers in the edge detection model to obtain a plurality of feature maps; performing Hadamard product matrix operation on the initial detection image and the plurality of feature maps to obtain a plurality of edge detection result images; and inputting the plurality of edge detection result images into a trained segmentation model for segmentation calculation to obtain a final output segmentation image. According to the method and system, the trained edge detection model is used for boundary segmentation assistance, the training difficulty is reduced, the refining degree and accuracy of edge segmentation are improved, an interested target area can be segmented in a complex road environment, and effective information is extracted.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a traffic road image segmentation method, system and computer-readable storage medium based on edge detection. Background technique [0002] With the expansion and improvement of the transportation network and the development of artificial intelligence, smart transportation has emerged and developed rapidly. Most of the data information of intelligent transportation comes from the images collected by cameras, so image analysis and understanding are extremely important in intelligent transportation. Some image information acquisition requires relatively fine-grained processing, such as lane line detection and lane segmentation, road water detection, road marking damage detection, etc. These scenarios are where the segmentation algorithm comes into play. [0003] For image analysis and understanding, it has gone through three processes of classification-detection-...

AI Technical Summary

Problems solved by technology

[0004] The traditional image segmentation technology has low precision and poor real-time performance, and the content of the segmented image is relatively rough, which is far from the point of practical application.

In recent years, the accuracy of the segmentation algorithm based on deep learning has been greatly improved, and the real-time performance is also gradually improving. However, it is still insufficient to be used in actual scenes. The targets that can be segmented are limited. In order to segment multiple targets, multiple targets are required. Segmentation process, it is impossible to segment all the targets of interest on an image as much as possible at one time

[0005] In addition, in the traffic road scene, due to the particularity of the camera position and the geographical location of the road, the interference of background information is large, the position of the target of interest is limited and the amount of information is small, which makes segmentation difficult

At present, most of the segmentation algorithms based on deep learning are aimed at ideal scenes, with a wide range of targets of interest and small background interference, which means that they are not adaptable to complex environments.

Images