Behavior Recognition Method of Motion Trajectory Based on 3d Stationary Wavelet

Abstract

The invention discloses a behavior recognition method based on a 3D stationary wavelet transform trajectory, which mainly solves the problems of insufficient time-space information extraction and low robustness of the existing wavelet domain behavior recognition technology. The technical solution is: 1. Decompose the video with time-space separable 3D stationary wavelet transform to obtain the high-frequency and intermediate-frequency subbands in the time domain and subbands in each direction in the space-time domain; 2. Based on the high-frequency and intermediate-frequency subbands in the time domain 3. Extract the feature points in the fused subbands based on the energy threshold; 4. Use the space-time domain subbands in each direction to construct the wavelet coefficient descriptors of the feature points, and according to their Euclidean distance in the adjacent Match the feature points between frames to get the motion trajectory; 5. Construct the wavelet direction energy histogram feature around the trajectory, and build the bag-of-words model of the histogram feature, and then use the SVM classifier to identify and classify. The invention improves the accuracy rate of human behavior recognition and can be applied to abnormal behavior detection and human-computer interaction.

Description

technical field [0001] The invention belongs to the technical field of video processing, and further relates to a behavior recognition method, which can be used for abnormal behavior detection and human-computer interaction. Background technique [0002] In recent years, computer vision has developed rapidly as an emerging discipline. Behavior recognition, as a key technology for video analysis and understanding, has important academic value, potential commercial value and huge application prospects, making it quickly become a research hotspot in the field of computer vision and Difficulties have been widely used in human-computer interaction fields such as video retrieval, intelligent monitoring, robot navigation, intelligent transportation, and game entertainment. More and more scholars and institutions have carried out a lot of research work in related fields. The key to human behavior analysis is to capture motion information in video and the relationship between frame s...

AI Technical Summary

Problems solved by technology

This method uses the generalized Gaussian distribution to fit the wavelet coefficients, which can obtain the space-time information of the video sequence and the dependence between the wavelet coefficients to a certain extent. The selection of probability model parameters as feature descriptors is helpful for feature dimensionality reduction, but only global The feature representation method is insufficient to obtain video structure information, and is sensitive to interference such as complex background and noise

[0005] The properties of the two-dimensional spatial domain and the one-dimensional temporal domain in video are very different, so it is intuitive that the two should be treated differently instead of just extending the two-dimensional spatial method to the joint three-dimensional space, along the Tracking interest points along video sequences is a method suitable for dealing with the above problems found by scholars in recent years. However, no scholars have used wavelet transform to extract motion trajectories in videos, so as to introduce the advantages of trajectories into wavelet domain behavior recognition.

Images