Video-Based Fall Detection System Adapting To Different Body Shapes

In daily life,warning signs of ‘Caution Wet Floor' can be seen everywhere.These signs remind people to guard against falling down.Whether at home or in nursing homes,escalators in public areas,or factories with poor working conditions,falling down threats people of all ages.It will be difficult to save yourself when you fall down seriously.Therefore,the study of automatic fall detection system is of great significance to people's health and safety.We investigate existing wearable,environmental and video-based automatic fall detection solutions so that people can be help timely after falling.Video-based fall detection solution is chosen to design the system because of the rich falling down characteristics and high detection freedom.In the system design,we are devoted to solving the problems of large volume and poor real-time in fall detection system.In consideration of the computing power of ARM,DSP and FPGA,the video-based embedded system for fall detection is built.This system is composed of FPGA,camera and alarm circuits.FPGA is the control core.Camera is used to acquire images.Alarm circuits involve acousto-optic alarm,message alarm and voice alarm.In algorithm design,the accuracy of fall detection is influenced by artificial experience too much.The error rate increases obviously due to the body shapes changing.We propose a fall detection algorithm with body shape adaptive.First of all,extracting human image by background subtraction.Secondly,calculating the fall characteristics uses minimum rectangle method.Then the fall thresholds are set by decision tree.Moreover,different body shapes are distinguished by Euclidean distance.According to the grade of the sensitivity of the different fall characteristics,the fall detection is realized under the high accuracy rate.In the function design of FPGA,we adopt top-down design idea and utilize Verilog as system logic description method.The system modules are divided into different function.They involve camera driver,memory and peripheral alarm circuits.And loading the body shape adaptive fall detection algorithm.The experimental results show that the average accuracy of the fall detection algorithm is 95.2%.Also,being compared with the combination of artificial experience threshold,the average accuracy rate is increased by more than 15%,which can ensure a high accuracy when there are large differences in the body shape.Under the condition of 640×480 resolution and 30 fps frame rate,the average response time of voice alarm and acousto-optic alarm is less than 2 seconds,and the average delay of sending alarm message is no more than 5 seconds.This system can give an alarm immediately after people falling.