Old people abnormal behavior identification method

Abstract

The invention discloses an old people abnormal behavior identification method, which relates to image information processing of image feature extraction, and is based on an improved and trained stack-type convolution ISA model. The old people abnormal behavior identification method comprises the steps of: establishing a video sample database of old people behavior patterns and partitioning a video sample; preprocessing video sample data; extracting temporal and spatial characteristics from the video sample data by utilizing the improved and trained stack-type convolution ISA model; clustering the temporal and spatial characteristics to obtain a visual word list and a video visual word frequency histogram; training an X<2> kernel support vector machine SVM classifier model; and identifying old people abnormal behaviors. The old people abnormal behavior identification method does not need to carry out human body segmentation and background modeling, directly extracts the temporal and spatial characteristics from the video data through establishing an old people behavior database and adopting an unsupervised learning method, realizes the identification of the old people abnormal behaviors, and overcomes various defects in the prior art.

Description

technical field [0001] The technical solution of the present invention relates to image information processing of image feature extraction, specifically a method for identifying abnormal behavior of the elderly. Background technique [0002] In the field of computer vision technology, the identification of abnormal behavior of the elderly is through the real-time monitoring and intelligent services of the elderly living alone through the computer system. When in an abnormal behavior state, the computer system can detect and alarm in time, so that the elderly can be rescued in time. At present, the research on the identification method of the abnormal behavior of the elderly is a research hotspot, which has a lot of room for development. In the prior art: CN104850841A discloses a method for monitoring abnormal behavior of the elderly combined with RFID and video recognition. The identity information of the elderly is identified by wearing an RFID tag for the elderly, and the...

AI Technical Summary

Problems solved by technology

CN102799873A discloses a human body abnormal behavior recognition method, without the need for human body segmentation and background modeling, through clustering the spatio-temporal features of the intense human body movement area, the movement modeling is realized, and the recognition of human body abnormal behavior is realized. Abnormal behavior templates classify behaviors, but the features extracted by this method are artificially designed features, which are not necessarily applicable to all behavior databases, and have great limitations

Quoc V.Le proposed the so-called independent subspace analysis (ISA) feature extraction method in his paper "Learning hierarchical invariantspatio-temporal features for action recognition with independent subspace analysis", which uses unsupervised learning methods Extract spatio-temporal features directly from video data to complete behavior recognition. This method has achieved good results in several mainstream human behavior databases. Larger, iterative convergence speed is slow

Images