Human being tumble monitoring method in video monitoring system

Abstract

The present invention provides a human being tumble monitoring method in a video monitoring system. The method comprises: employing a Vibe algorithm to extract prospect information, extracting 10-dimensional feature data such as Hu distance features, a center changing rate, a human body depth-width ratio, an effective area ratio and the like aiming at the prospect information, employing an Adaboost algorithm to perform training of feature data extracted in the samples, obtaining a strong classifier, and employing the strong classifier to process real-time images in the video monitoring system to detect whether there is a tumble condition or not. According to the experiment evidence, compared to other tumble monitoring methods, the human being tumble monitoring method in the video monitoring system has the detection accuracy which can reaching up to 93% in the condition of improving the operation speed and reducing the memory occupation so as to improve the condition of low identification accuracy and bad timeliness in the current intelligent monitoring system.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a method for monitoring human falls in a video monitoring system. [0002] technical background [0003] In recent years, as our country enters an "aging" society, caring for the elderly, and the care of the empty-nest elderly have become increasingly prominent. It has become an important proposition for scientific development to let the elderly "have something to depend on and support when they are old". Because the elderly often suffer from sudden fatal diseases, once they faint due to sudden disease and other circumstances, if they are not discovered in time, it is very likely to threaten their lives and cause irreparable regrets to their families. In today's rapid development, available human resources are becoming more and more scarce, and the care of the elderly has become an urgent social problem. [0004] At present, the existing detection methods for human body posture are...

AI Technical Summary

Problems solved by technology

The direct classification method often does not consider the temporal relationship, and only directly learns and trains its behavioral features for a single frame of images. The template matching method stores known behavioral templates in the standard behavioral template library, and extracts them from the behavior recognition process. The template is compared with the existing template for behavior classification identification and classification. The template matching method has a simple calculation process, but its robustness is poor, and it is greatly affected by the standard behavior template library.

The state-space law fully considers the dynamic process of human movement, and defines each posture as a collection of several states. The mutual switching between states is described by a probabilistic statistical model. The most commonly used state-space method is the hidden Markov model. (HMM), but the state space method includes a large number of iterative operations, the computational complexity is high, and it is not suitable for real-time monitoring occasions

Images