Distributed Fall Detection Combined With Age Estimation

With the increasing aging of the elderly,the health monitoring of the elderly such as falling suddenly has become a social problem.The fall detection method based on computer vision provides an effective solution for the health monitoring of the elderly.However,there are still some problems to be solved.(1)the existing research often ignores the analysis of the fall priority of the crowd,and only focuses on the results.However,the consequences of falls may be completely different among people of different ages;(2)most of the previous fall detection methods are centralized,poor real-time and scalability,which cannot give priority to the detection of key groups and lack a fall pre-judgment mechanism.As for the problem(1)that the existing fall detection methods cannot detect pedestrian falls based on their age groups,this dissertation combines age estimation and fall detection algorithms and proposes an age estimation algorithm based on Soft Stage Regression-Shallow(SSR-S)network.The age estimation algorithm extracts age-related features of the face image and designs a multi-granular age estimation method based on regression calculation,an accurate estimation of the age of pedestrians is achieved.As for the problem(2)that the existing fall detection methods have poor real-time and scalability,this dissertation proposes a fall detection model that combines motion feature values and Shallow-Convolutional Neural Networks(S-CNN)under a distributed architecture.In this model,ellipses are applied to fit the head and trunk of the human body respectively,and the motion characteristic values of the human body are extracted.According to the vertical velocity of the head ellipse,the threshold of fall pre-judgment is constructed to realize the pre-judgment of pedestrians under 60 years old.Finally,S-CNN model is applied for more accurate fall detection.In addition,a distributed model based on hierarchical agents is proposed,and a priority scheduling algorithm based on age estimation is proposed for this model.This algorithm can prioritize key tasks according to the priority of the task when some servers are congested or even down,reducing queuing delay and shortening the execution time of task.In order to verify the effectiveness of the algorithm,this dissertation has conducted extensive experiments.Experimental results show that compared with other latest methods,the age estimation algorithm based on SSR-S network has the lowest MAE,which can reach 7.59.The fall detection model based on motion feature values and SCNN can effectively prejudge fall behaviors and daily behaviors,and achieve a 0%missed detection rate of fall behaviors.The accuracy rate of fall detection methods based on S-CNN can reach to 90.5%.In addition,compared with the traditional centralized model,the distributed model improves the efficiency of task processing.Among them,the priority scheduling algorithm based on age estimation can solve the problem that the server cannot prioritize the detection of key tasks when some servers are congested or even down due to unexpected circumstances,and it is more stable than the round-robin scheduling algorithm.