Human motion tracking method based on distributed collaborative learning

Abstract

The invention discloses a human body motion tracking method based on distributive collaborative learning for mainly solving the problem that the hardware needed by the existing technical training has high cost and long training time and is powerless to large data set. The tracking method is implemented by the following steps of: (1) dividing a video into frame images and extracting human body position block diagrams from the images; (2) extracting human body characteristics in the block diagrams by using a descriptor; (3) mapping the extracted characteristics into a space consisting of mapping vectors by a random characteristic mapping method; (4) forming a human body motion tracking model by mapping vectors and actual postures of a training sample, and dividing the human body motion tracking model into a plurality of sub-models; and (5) collaboratively working out a common solution of a plurality of sub-models by using a plurality of learning machines, and estimating the actual motion postures of the test sample by utilizing the common solution. Compared with the traditional human body motion tracking method, the tracking method disclosed by the invention has the advantages of low hardware cost and short training time on the premise of achieving the same precision, and can be used for motion capture, man-machine interaction and video monitoring.

Description

technical field [0001] The invention belongs to the technical field of computer vision and video image processing, mainly relates to video human body motion tracking and three-dimensional attitude recovery, and can be used for motion capture, human-computer interaction and video monitoring. Background technique [0002] Video human motion tracking is one of the major hotspots in the field of computer vision in the past two decades. People are the core content and reflect the core semantic features of images. This type of technology has been initially applied in many fields such as motion capture, human-computer interaction, and video surveillance, and has great application prospects. The understanding and interpretation of video human motion tracking belongs to the category of video image processing, and also involves many disciplines such as pattern recognition, machine learning and signal processing. A series of researches on 3D human motion tracking and posture recovery ...

AI Technical Summary

Problems solved by technology

However, most of the modeling methods are completed by a single learning machine, which makes the memory requirements and computing time very large, especially today when the data to be processed grows geometrically, these methods seem powerless, and their practical applications are greatly limited.

Images