Detection Method Of Human Fall Based On Infrared Image Features

With the progress of society and the development of science and technology,people's living environment and social medical security have been greatly improved,which makes the life span of human beings have a longer extension than before,and the health care problem of the elderly is also prominent.Fall is the main cause of the elderly's accidental injury.Many elderly people are disabled or even died due to fall,so it is very important to study the fall detection technology.In recent years,most of the fall detection based on computer vision uses visible light image,which results in the poor detection effect of the fall detection system in the light conditions or night environment,and can not work.Infrared image is not affected by the light,has strong anti-interference,and can work all day.Therefore,an infrared image detection method for human body falls is proposed.The infrared image of human body falls is obtained by infrared camera,and then the key point based target detection method is used to detect human body falls.The research content of this paper is as follows:Firstly,the characteristics and preprocessing methods of infrared image are studied.Combined with the imaging characteristics of the infrared camera used in this paper,the infrared image is preprocessed,such as denoising and linear stretching.The results show that the infrared image quality is good,the target is clear without too much noise,which lays the foundation for the later fall detection.Secondly,this paper studies the public data set of human body falls,and studies its shooting methods and contents.It is found that the current fall data set images are all single person,and the fall posture is generally forward and backward or side fall without shelter,which is relatively single.In view of the above problems,this paper uses infrared camera to establish the scene of human body falling,records the infrared image of human body falling data,increases the kneeling and leg lifting posture,increases the number of people and interference objects such as tables and chairs,and increases the scene of different degrees of occlusion when falling,so as to establish their own falling data set.Finally,in order to improve the accuracy and real-time of fall detection,this paper proposes a key point based fall detection method.The Center Net target detection network is used to train and test the network by using the established infrared image falling data set,and the comparison experiment with Yolo V3 is carried out.The experimental results show that the method proposed in this paper can achieve more than 99% accuracy on the self built data set,and the performance indexes in all aspects are better than Yolo v3.