Scroll detection method, device and apparatus

Abstract

The embodiment of the invention provides a banner detection method, device and apparatus. The method comprises: obtaining a to-be-detected image, and inputting the to-be-detected image into a preset deep learning model, wherein the preset deep learning model is a deep learning model obtained by training a plurality of collected sample images containing banner objects in advance; obtaining to-be-detected objects output after the preset deep learning model processes the to-be-detected image and the category confidence of each to-be-detected object, wherein the to-be-detected object is extractedfrom the to-be-detected image by the preset deep learning model; for each to-be-tested object, judging whether the category confidence of the to-be-tested object meets a preset banner category confidence condition or not; and if so, determining that the to-be-detected object is a banner. Compared with a mode of detecting the banner from the to-be-detected image based on color recognition and corner detection, the method can detect the shielded banner or the banner of which the color is close to the background color, and improves the detection accuracy of the banner in the image.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a banner detection method, device and equipment. Background technique [0002] With the development of image processing technology, the image processing technology is widely used to detect whether there is a target object in the captured image. For example, image processing techniques are used to detect banners in acquired images. [0003] Typically, target objects in an image can be detected based on color recognition and corner detection. For example, when detecting banners in a captured image, a color histogram of the image may be acquired to determine whether the color histogram is a bimodal image. If yes, there may be banding in the image, indicating that there are objects in the image that have a high contrast color to the background color of the scene being captured. Then select an area corresponding to the peak of the color histogram in the image, and perform ...

AI Technical Summary

Problems solved by technology

[0004] However, if the target object in the image does not have a high-contrast color with the background, it will not be possible to use color recognition to distinguish the target object from the background, resulting in inaccurate color recognition results

For example, when the background is a white wall and the banner is also white, the color discrimination is not high, resulting in the detection result may be that there is no banner in the image, and the recognition result is inaccurate

Or when the target object is occluded, the occlusion changes the shape of the target object, which will lead to inaccurate corner detection results

For example, if one of the top corners of the banner of the target object is covered, the covered banner is no longer a rectangle. When performing corner detection, it will be judged that the banner of the target object is non-rectangular, there is no banner in the image, and the recognition result is inaccurate

[0005] Therefore, when detecting banners in an image based on color recognition and corner detection, the detection accuracy is not high

Images