A method of identifying motion of local matching window based on sliding window

A local matching and sliding window technology, applied in the field of video recognition, can solve the problems of inability to recognize similar actions and low recognition rate, and achieve the effect of improving action recognition rate, improving representation, and increasing time constraints
CN104408461AActive Publication Date: 2015-03-11ZHEJIANG UNIV OF TECH
3 Cites 12 Cited by

Patent Information

Authority / Receiving Office
CN ยท China
Current Assignee / Owner
ZHEJIANG UNIV OF TECH
Publication Date
2015-03-11

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

A method of identifying a motion of a local matching window based on a sliding window includes the following steps: 1) obtaining a depth image sequence of a human in a scene from a stereo camera, extracting a position of a 3D pitch point from the depth image, and using a 3D displacement difference between postures as a feature expression for each frame of depth image; 2) studying descriptors in a training set using a cluster method to obtain a feature set, and using the feature set to perform feature expression on each descriptor, thereby obtaining an encoding expression of each image; 3) performing motion segment division on the whole motion image sequence using a local matching model based on the sliding window, and obtaining a feature histogram expression of each motion segment; and 4) connecting the feature histogram expressions of all segments in series using a long vector to obtain the feature expression of the entire motion. The method effectively identifies similar motions, and the rate of identifying is extremely high.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention relates to the field of video recognition, in particular to an action recognition method. Background technique

[0002] Human motion feature description is the expression of human body posture information in sequence images, and it is an important part of human action recognition. Human body motion is a chained non-rigid body motion, that is, the motion of each part of the body is a rigid body motion, and from the overall point of view, the motion of the human body is highly nonlinear and non-rigid.

[0003] Human action recognition is mainly divided into spatiotemporal-based action recognition methods and sequence-based action recognition methods. First construct a corresponding three-dimensional X-Y-T model for each action in the training video set, and then determine the type of the test action by matching the action sequence to be recognized with the three-dimensional X-Y-T model of each action in the training set. Time-space-based ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More