Human body movement tracking method based on bidirectional Markov Monte Carlo particle filter

Abstract

The invention discloses a human body movement tracking method based on a bidirectional Markov Monte Carlo particle filter. The method comprises the steps of building video time-space body models, andextracting model characteristics of the video time-space body models; selecting a frame image in a target video time-space body, performing bidirectional particle filtering on the image through the bidirectional Markov Monte Carlo particle filter to obtain a group of solutions, matching the obtained solutions with the model characteristics of the video time-space body models, and predicting a movement type according to a matched optimal solution; and feeding back the predicted movement type to an adjacent frame image for updating a movement type of the adjacent frame image, then returning to continue filtering, and circulating the process until all images of the target video time-space body are subjected to particle filtering. The movement type and a movement state are predicted together,and tracking and identification are performed at the same time, so that a good tracking result and an accurate identification result are obtained.

Description

technical field [0001] The present invention relates to the technical field of human body motion tracking, in particular to a human body motion tracking method based on a two-way Markov Monte Carlo particle filter. Background technique [0002] Human motion tracking is currently a research hotspot in the field of artificial intelligence and computer vision. In recent years, many tracking methods, especially Particle Filter (PF), have achieved good tracking performance. But in the particle filter system, the accuracy of the posterior probability density depends on the effectiveness of the particles, and a very effective way to improve the efficiency of particle effectiveness is to conduct a global search. However, global search will bring a very large amount of calculation and increase the running time. [0003] In order to solve this problem, the more effective method currently used is to use the human motion prediction model. With the guidance of the human motion model, ...

AI Technical Summary

Problems solved by technology

[0005] However, in traditional tracking methods, the identification process is not performed, or the identification process is performed after the tracking process

Images