A hand recognition method for supermarket shopping cart based on sobel operator

Abstract

The invention discloses a hand recognition method of a supermarket shopping cart based on a Sobel operator, which comprises the following steps: firstly, a 2D camera is used to obtain a color image, and a pre-made template image is used as a mask; secondly, the image is grayscaled , use the Sobel operator to perform convolution operation on the grayscaled image, then perform binarization processing on the processing result, and perform mask operation with the color image to obtain the ROI image; finally, perform hand-held product judgment on the ROI image, The hand-held product judgment includes a skin color modeling part, a hand-held product judgment part and a statistical part; the present invention adopts a Sobel operator, which weights the influence of the position of the pixel, which can reduce the edge blur, so the effect is better; Compared with sub-positioning, it has strong anti-noise ability; because the Sobel operator is in the form of a filter operator, when used to extract edges, it can use a fast convolution function, which is simple and effective; the amount of calculation is small, and it can be used in real-time systems.

Description

technical field [0001] The invention relates to an image recognition method, in particular to a Sobel operator-based hand recognition method for a shopping cart in a supermarket. Background technique [0002] In the shopping cart image algorithm, we use a depth camera to segment hand information; according to different hardware implementation methods, there are currently three mainstream 3D machine visions used in the industry: structured light, TOF time light, and binocular stereo imaging. This method can reconstruct the image well in 3D, so as to accurately segment the image. However, due to the limitation of the principle, there are detection blind spots for reflective, transparent, light-absorbing and other objects; moreover, due to the occlusion problem of one viewing angle, it is impossible to accurately judge empty-handed or hand-held products. Therefore, Multiple perspectives are desperately needed. [0003] The applicant previously applied for an invention patent n...

AI Technical Summary

Problems solved by technology

On this basis, the applicant proposes to collect images simultaneously from multiple perspectives to further improve the accuracy of hand feature extraction; since 3D cameras are expensive, it is not easy to install multiple cameras to expand the perspective, and considering the second perspective as an auxiliary judgment, the cost For lower 2D cameras, it is difficult to extract the foreground image from the images acquired by the 2D camera. If algorithms based on motion detection such as frame difference method and background elimination method are used, the frame difference method will lead to inaccurate foreground extraction, and background elimination will be affected. The impact of lighting is large, so an algorithm that does not rely on motion detection is urgently needed

Images