Target distribution law-based abnormal human behavior detection method

Abstract

The invention relates to a target distribution law-based abnormal human behavior detection method. A pre-trained depth learning network is used, according to deployment scene target learning trainingfine adjustment parameters, target detection is carried out, and the states of a person and an object in the scene are detected; according to a set state abnormal alarming point, abnormal alarming istriggered and carried out; the real-time positions of the person and the object in an area are acquired at the same time; the position information is returned; an area target point is set, the distance between the person position and the target point is calculated, the distance between the object position and the target point is calculated, the distances between the person and various objects arecalculated, the calculation results form curves, an object distribution activity curve, a person state activity curve and a curve of the distance between the person and a preset marking item are formed respectively, and a long-term iteration regression curve is acquired; and when the curve fluctuation exceeds a preset deviation range and does not satisfy long-term law and special date law curves,an abnormity is triggered for alarming and reminding.

Description

technical field [0001] The invention relates to a method for detecting abnormal behavior of personnel based on target distribution rules, and belongs to the technical field of health protection. Background technique [0002] With the increasingly serious problem of population aging, the demand for private guardians is increasing year by year. With the current level of social development, the level of medical resources and the number of guardians are limited, it is difficult to provide a full range of personalized guardianship services for each need. handling may lead to serious consequences. [0003] Most of the existing monitoring systems for the elderly and patients who need to be cared for use contact sensors to collect physiological parameters such as the heart rate, pulse, and exercise of the elderly, and pressure sensors to collect postures such as lying and sitting. There are many types of sensors and complex system deployment. The wearable mode leads to poor user e...

AI Technical Summary

Problems solved by technology

With the current level of social development, the level of medical resources and the number of guardians are limited, it is difficult to provide a full range of personalized guardianship services for each need. treatment, which may lead to serious consequences

[0003] Most of the existing monitoring systems for the elderly and patients who need to be cared for use contact sensors to collect physiological parameters such as the heart rate, pulse, and exercise of the elderly, and pressure sensors to collect postures such as lying and sitting. There are many types of sensors and complex system deployment. Wearable style leads to poor user experience

[0004] In the existing technology, the position acquisition mostly uses active infrared and passive infrared sensors, which are easily affected by occlusion, are not accurate enough and require a large number of layouts; in the prior art, some abnormal feedbacks are used in the time series of sensor feedback in big data The position of the person trains a Bayesian network. The Bayesian network model only relies on location variables, time variables and abnormal probability variables to model. The model features are too few and the dependencies between features are too large. The performance of the established Bayesian network model is very poor. Poor, leading to overfitting of the model, lack of generalization ability and individual uniqueness, and judging whether the ward has abnormalities is a complex multi-dimensional problem. Only relying on location and time to speculate whether there are abnormalities has a high rate of misjudgment

Images