Video image sequence segmentation system and method

Abstract

The present invention provides a video image sequence segmentation system. The system includes a position estimator, a classifier and a motion state segmenter; the position estimator is configured toreceive the video image sequence of an object, wherein the video image sequence contains a plurality of consecutive frames indicating different motion states of the object; and the position estimatoris further configured to determine the positions of a plurality of body key portions of the object in each frame of the video image sequence on the basis of a deep learning algorithm, wherein the deeplearning algorithm includes an error iterative feedback algorithm or a partial association field algorithm; the classifier is configured to determine parameters indicating the types of the motion states of the object in each frame according to the positions of the plurality of body key portions; and the motion state segmenter is configured to segment the video image sequence on the basis of the parameters so as to obtain corresponding frames corresponding to each motion state of the object. With the system adopted, a video image sequences acquired in a non-strictly-controlled environment canbe segmented, and therefore, the remote monitoring of the motion capability of the object can be realized.

Description

technical field [0001] The present invention relates to video image sequence segmentation, in particular to a system and method for segmenting video image sequences corresponding to different motion states of objects in the process of evaluating the action ability of objects. Background technique [0002] Mobility monitoring is of great interest for people at high risk of falls, such as the elderly, or those with Parkinson's disease. The Timed Up-and-Go (TUG) test is a widely accepted test for assessing the mobility of subjects such as elderly people, and it typically involves the subject going from sitting in a chair, getting up, walking forward, turning around, walking back The whole process from sitting back to the chair specifically involves the movement stages of several different motion states of "sit", "sit to stand", "walk", "turn around", "walk back" and "sit back". The duration of each motor phase is a very meaningful parameter for evaluating the mobility of the s...

AI Technical Summary

Problems solved by technology

This method is able to obtain relatively accurate test results, but for non-professionals, the construction and setup of the device is difficult, requires a lengthy calculation process, and attaching the device to the subject may cause discomfort to the subject, especially for those with limited mobility. object

In the second type of method, environmental sensors such as force sensors and laser ranging scanners are used to measure and perform motion state recognition based on the measurement results. This method does not require equipment attached to the subject, but the installation and calibration of these Environmental sensors remain challenging for non-professionals

[0004] In the third type of method, motion state recognition based on video images, usually first obtain the outline of the object, and then identify the motion state of the object with the help of features such as height, width, and aspect ratio, which requires Obtaining video of objects in the environment, for example, requires a consistent background and a fixed camera angle. Due to the limitations of video analysis methods, the results are not satisfactory

In the fourth type of method, a depth camera is used to capture additional 3D information, this kind of device is not common in daily life and is not as easy to operate and set up as a 2D camera for non-professionals

Images