Mood recognition method based on pet postures

Abstract

The invention relates to a mood recognition method based on pet postures. The mood recognition method comprises the steps of performing graying processing on an original image to obtain a grayed image; performing edge processing on the graying image based on a Sobel operator to obtain an edge contour image, processing the edge contour image to obtain a maximum enclosing rectangle, and then cutting out the foreground target from the original image to obtain a preprocessed image; processing and fusing the pre-processed image and the original image by using different feature extraction networks, and predicting to obtain a pet category; performing contour filling on the preprocessed image to obtain a mask image, sequentially corroding 8-pixel neighborhoods of contour points of the mask image to obtain skeleton features, and further calculating to obtain an included angle between a tail node and a body node; determining pet mood based on the included angle and the category; the skeleton is obtained by judging and processing each pixel point of the mask image, so that a user does not need to worry about that the tail cannot be separated from the body, the included angle between the tail and the body is easy to calculate, and pet classification identification is added for the accuracy of identification.

Description

technical field [0001] The present invention relates to the field of computer vision. Specifically, it relates to a mood recognition method based on pet gestures. Background technique [0002] Because the company of pets can bring warmth, joy, and physical and mental changes to people, the number of pets keeps increasing, and people treat them as family members, but due to language barriers, there are barriers to communication with pets, so we only Be able to judge the mood of pets from their behavior. [0003] In the prior art, based on deep learning and SVM for animal mood recognition, thirteen key points of the body of a large number of animal pictures are marked, and a deep learning animal pose estimation model is constructed, so that the model can estimate the center point position and the position of the animal. The positions of the thirteen body key points, based on the position change of the animal's center point, the position relationship and relative position cha...

AI Technical Summary

Problems solved by technology

[0005] In digital image processing technology, because it only classifies according to the pixel features of the image, the classification robustness is low, but the sensitivity of digital image processing methods to pixels is very significant; in deep learning technology, due to the feature extraction network based on various structures It can automatically perform feature extraction, avoid manual participation in the process, improve classification accuracy, and reduce labor costs. However, the feature extraction process is "black box" and relies heavily on training data sets, which has also become a drawback of deep learning technology.

Most of the images faced by the existing image classification tasks are images with complex scenes and chaotic backgrounds. How to extract typical features from them is a difficult problem that the above two methods need to face together.

Images