Seaman tumble detection method and device based on improved Blazepose-LSTM

The invention provides a sailor falling detection method and a sailor falling detection device based on an improved Blazepose-LSTM (Long Short Term Memory). According to the detection method, human body key point information is extracted through a Blazepose human body key point information extraction network, a rigid feature detector module in the human body key point information is optimized through an offset vector, and crew falling judgment is carried out through a long-short-term memory neural network. Human body key point information is automatically extracted from a video image acquired by a visual sensor, internal data association features are analyzed, the process from RGB camera image processing to sailor falling detection is achieved, the detection accuracy and real-time performance of a sailor falling detection algorithm are improved, the effects of low computing power and real-time detection are met, and the sailor falling detection efficiency is improved. And the practicability and the strong gen