一种结合Alphapose和LSTM的人体摔倒检测模型

Many Human-Body-Fall-Dow n Detection models are faced with problems like lower adaptability and higher w rong-detection rate in different detection scenes，and targeting on these shortcomings this research proposes a Human-Body-Fall-Dow n Detection Model based on the human skeleton keypoints and the LSTM neural network． This model detects the skeleton keypoints of the continuous multi-frame human body by Alphapose，and then divides the coordinate sequences of the skeleton keypoints into X and Y coordinate sequence，and then inputs them respectively into LSTM neural network to extract the time-order character; the last step is to input the LSTM hidden layer output vector into a full connection layer to obtain the results． This research uses public data set M uHAViMAS and Le2i to execute this experiment，and compares itself with many other detection models． The results show that this model has relatively high detection accuracy in multiple scenes，multiple view s，and multiple poses of falling．