The Design Of The Real-time Detection System For Indoor Falls Of The Elderly Based On ZYNQ Platform

With the rapid development of our country's economy.At present,China's population aging problem is increasingly serious,which causes an increasing number of elderly people being cared for by services such as nursing homes.Caring for the elderly is becoming increasingly difficult,hence,there is an urgent need to find a way to solve this problem.In the nursing process,the elderly people have obvious physical degeneration due to their age,it is easy for the elderly to fall.According to statistics,the most frequent accidental injuries of the elderly are indoor falls,if the oldderly is not found to fall in time,they is likely to be in danger of life.In order to timely detect the accidental fall of the elderly in the room,there are many automatic detection and automatic alarm systems at home and abroad.For example,there is a wearable fall detection system,which collects movement information of human body through acceleration sensors and gyroscope sensors.When the elderly suddenly fall,the system will send out an alarm.Wearable fall detection system can better detect whether the elderly have fallen and send out an alarm.However,wearable devices may increase the burden on the elderly and cause physical discomfort to some extent.In addition,the system is not suitable for use in places such as bathrooms,and the bathroom is the place that the old person happens to fall more easily.And when the elderly are exercising,which may lead to system misinformation.Because of the problems of the wearable detection system,a fall detection system based on environmental awareness was proposed.The system can obtain and transmit the behavior information of the elderly in the current environment in real time through sound sensor,pressure sensor,infrared sensor and wireless sensor network,the system will not increase the physical burden on the elderly.The system can better determine the elderly fall.But the system is vulnerable to interference from the environment,such as when the object falls to the ground,it will trigger a false alarm system.In addition,if the system wants to cover the whole range of indoor activities of the elderly,it is difficult to install and expensive.Because of the problems of the fall detection system based on environmental awareness,the real-time detection system of the elderly's fall based on video surveillance was proposed.The system uses a camera to capture image of the elderly in real time and transmit it to the central control room.Then,human eye recognition or image processing technology is used to determine whether the elderly have fallen down in real time.This method can accurately determine whether the old man falls,without adding extra burden to the old man's body.The installation convenience and cost is lower.However,the scheme based on video surveillance may reveal the privacy of the elderly and increase their psychological burden.In this reashearch,aiming at the problems of the above-mentioned fall detection system for the elderly,this paper proposes a real-time fall detection system for the elderly based on ZYNQ(ARM + FPGA)platform.This design aims at more efficient and high-quality detection of the elderly fall,for the elderly for the golden treatment time.This system adopts the video surveillance scheme and solves the problems of violating the privacy of the elderly and increasing the psychological burden of the elderly.The system uses ZYNQ hardware platform to monitor the elderly's fall in real time.When the elderly are detected to fall,the alarm function of the Labview upper computer buzzer will be triggered,and the nurse will be informed of the alarm in time.In the realization of the recognition of the elderly fall,this paper mainly makes the following five aspects of work:1.In this system,the convolution neural network algorithm in the field of deep learning is used to identify the images of elderly people who fall and those who do not fall.Firstly,the daily fall posture of the elderly was analyzed,then collect pictures of the old man when he fell and moved normally,forming the training set of convolutional neural network.2.The network structure was built in Caffe deep learning platform,and the convolutional neural network model was trained with training sets,so that the network model could identify whether the elderly had fallen down,and the script was used to extract the weight value and bias value in the network model.3.Set up the system hardware platform for acquiring videos based on ZYNQ.Zynq-7000's unique "ARM+FPGA" framework was used to design the overall framework of the system,and IMX222 camera was used to obtain image information.4.In order to protect the privacy of the elderly in video monitoring,the system will not upload the detected image data of the elderly to the PC or the cloud for processing.Instead,the convolutional neural network is deployed in the ZYNQ hardware by using SDSoC development software,and the acquired image data is directly processed on the ZYNQ hardware platform.And the identification results is uploaded to the Labview host computer for discrimination,if an elderly person is detected to fall,an alarm will be triggered.5.In order to achieve real-time performance of the system,the system analyzes the identification time.In this paper,pure software implementation scheme(ARM implementation scheme),hardware acceleration technology implementation scheme(ARM+FPGA scheme)and hardware acceleration optimization technology implementation scheme(ARM+FPGA optimization scheme)are used for comparison.This paper compares the time performance and hardware consumption of these three schemes,and finally selects the implementation scheme using hardware accelerated optimization technology,which optimizes the system from the bottom of the hardware.Finally,through the system test,the system can be efficient and high quality work.The recognition accuracy of the system is about 92%,and the recognition time is only 413 ms,which meets the real-time requirement of the system.The upper computer alarm display function is basically 0 delay,and the upper computer can timely inform the nursing staff without revealing the privacy of the elderly,which can better achieve the expected design purpose.According to the experimental results,this design can meet the needs of the indoor fall detection system for the elderly.The system can be well applied in nursing homes,families,hospitals and other places,can ensure the safety of the elderly with high efficiency and high quality.This system has a good feasibility through the field test.However,in later work,the system can be further optimized.Because in the system test,it is found that if the image collected by the low-definition camera,the system recognition accuracy will be reduced to a certain extent.The camera of this design finally chooses the HD camera of Sony to ensure the accuracy of system recognition.The cost is relatively high.In the future,this study may try to use image superresolution reconstruction technology,counter neural network and other technologies to reduce the system's requirements on image quality.to reduce the cost of the system.