An Improvement in Fall Detection System by Voting Strategy

The number of elderly people is increasing but the regenerative ability of those people is decreasing every day. Falls and the consequences of falls impact highly on their health. The fall detection systems are developed for years. This paper proposes the improvement in the fall detection system by the voting strategy. The three Kinect cameras provide the depth image to the fall detection systems. Each system performs (1) Frame Smoothing by Median Filtering, (2) Background Subtraction by Mixture of Gaussian Model and Binary Thresholding, and (3) Event Recognition to predict the fall or non-fall events by Convolutional Neural Network. The voting strategy results in the majority event from the three detections. The results show that the voting strategy increases the accuracy, precision, and fall detection rate to 96.48%, 90.20%, and 96.65% respectively.