Human body posture estimation method capable of adaptively configuring multiple models

Abstract

The invention discloses a human body posture estimation multi-model self-adaptive configuration method. The method comprises the following steps of: 1, predicting current video content parameters for high-precision human body posture estimation models of all sampling frames of previous t duration of a current processing video segment, and selecting a minimum configuration value required by the current processing video segment; 2, according to the lowest configuration parameter, performing fine granularity division on the sampling frames obtained by sampling to obtain each sub-region and a processing level, performing estimation processing by using the human body posture estimation model which is most matched with the processing level of each sub-region, and performing coordinate conversion to obtain a final estimation result; and 3, judging whether the current lowest configuration parameter is matched with a next video or not, if so, processing according to the steps 2-3, and otherwise, processing according to the steps 1-3 until the whole video to be processed is processed. The method is based on self-adaptive configuration of the multiple human body posture estimation models, and the execution efficiency of multiple-human-body-posture estimation can be improved under the conditions that calculation resources are limited and the precision of the calculation resources is not reduced.

Description

technical field [0001] The invention relates to the field of video analysis, in particular to a method for estimating human body poses with self-adaptive configuration of multiple models. Background technique [0002] At present, a large number of cameras are widely used in practice, such as in security scenes, various work scenes, and traffic scenes. With the popularization of camera applications, the work of video analysis is becoming more and more important, especially, the estimation of human body pose is a very important part of the function. [0003] At present, there are very mature human pose estimation methods, which can realize accurate detection of human pose in the case of a single person. However, when there are many people, the existing human pose estimation methods have at least the following limitations: frameworks with more accurate detection results (such as Openpose) require a lot of computing overhead, and the frame rate that can be processed per second ...

AI Technical Summary

Problems solved by technology

[0004] In view of the problems existing in the existing methods, the purpose of the present invention is to provide a human body pose estimation method with adaptive configuration of multiple models, which can solve the problems of the existing human body pose estimation methods, such as high computational overhead, slow processing speed or inaccurate results. The problem that the balance between processing speed and result accuracy cannot be achieved

Images