Method and device for realizing face and figure area selection based on 4K+AI

Abstract

The invention provides a method for realizing face and figure area selection based on 4K+AI. The method comprises the following steps: carrying out depth calculation of the resolution of a figure detection model through a head-shoulder algorithm, and obtaining a figure area; processing a to-be-processed image based on the figure area to obtain first and second coordinate information of a figure identification area, and obtaining third coordinate information of a figure identification amplification area by determining the front limit position of the figure area; and according to the first coordinate information, the second coordinate information and the third coordinate information, determining a figure identification amplification area. By adopting a face and figure area selection mode provided by the invention, the situation that an image is stiff to magnify and shrink, and the face is outside a screen or the basic shoulders and the head are not in the screen when the face is magnified and recognized can be effectively avoided; better figure areas and mistake proofing measures are optimized; a faster recognition response speed can be realized, and the use experience of a user is improved.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular, to a method and device for selecting human faces and human-shaped areas based on 4K+AI. Background technique [0002] Most of the current facial target recognition and humanoid target recognition support 1920*1080 resolution and are backward compatible; however, with the continuous improvement of camera resolution, there are more and more high-resolution large-scale images such as 2K and 4K. Accuracy also brings a huge amount of calculation; and once the resolution of the face detection model is smaller than the resolution of the image, it cannot be recognized. [0003] In response to this situation, the mainstream processing method is: uniformly segment the image to obtain a smaller-scale image, and then reduce the resolution for detection. Although such methods can achieve higher accuracy in object detection, they increase the computational cost. [0004...

AI Technical Summary

Problems solved by technology

However, the current AI face tracking algorithm still has the following defects: image zoom-in and zoom-out is relatively blunt, and the zoom-in recognition of the face may appear that the face is outside the screen or the basic shoulders and head are not in the screen.

Images