| With the rapid growing of the aging population in our country,research and development of products for the elderly in smart home have raised extensive attention.To ensure the safety of the elderly who is living alone,video camera or wearable sensors have been usually used to monitor the active status of the elderly.However,there are still some deficiencies in these two technologies.For example,the use of video camera brings the issue of privacy invasion,and the use of wearable sensors are relatively inconvenient thus are often forget to wear by the elderly.For these reasons,a new indoor personnel monitoring system based on low-resolution infrared array sensors is proposed for the elderly care in this dissertation.The main contributions of this dissertation can be summarized as follows:Firstly,the working principle and personnel activity status recognition algorithm for indoor personnel monitoring using low-resolution infrared array sensors are investigated in detail.It is found that using support vector machine method to recognize the posture of indoor personnel requires manual extraction of feature values which is very complex to extract the feature values.Then,the method of recognizing status of the persons in indoor environment using convolutional neural network is investigated.Secondly,a data acquisition system based on Panasonic's Grid-eye infrared array sensor is designed for collecting infrared thermal image in indoor environment.The system consists of a server and an acquisition device and they communicate with each other through WiFi module named RM04.The acquisition device is controlled by the STC12 series of the single-chip microcomputer which can realize temperature data collecting with a maximum rate of 10 frames per second and 64 pixels of data per frame.Then the collecting data is send to the server,where a data processing program is designed based on MATLAB and a GUI graphical user interface.Each data frame sent to the PC server by the acquisition device is converted into a pseudo-color image of 8×8 after processing,forming an infrared thermal image,which can be real-time displayed.Since the surface temperature of the human body is higher than other objects in the indoor environment,the active status of human can be monitored by observing the infrared thermal image.Finally,the use of convolutional neural network technology to process the collected infrared thermal images to monitor the changes in the active status of the target person is investigated.This method can automatically detect personal status of falling,standing and sitting in indoor environment.It is found that the indoor personnel attitude recognition model trained on the Caffe-based VGG16 model has a better effect with an accuracy of falling detection of 98.9%.Moreover,the effect of indoor background temperature and the wearing clothes on identification algorithms is further investigated.The results and conclusions of the above research provide a reference design scheme for indoor personnel monitoring systems based on infrared array sensors.Some new interpretations have been put forward on the personnel attitude recognition technology,which has important practical application value for development a personnel monitoring system used in smart home elderly care system. |
