The invention relates to the technical field of old people tumble detection, and discloses an old people tumble real-time detection
system based on attitude
estimation, which comprises a control terminal and a data
collection system, the data
collection system is in bidirectional
signal connection with the control terminal, and six daily attitudes including walking, standing, stooping, sitting down,
jumping and
lying down are systematically collected, so as to realize real-time detection of old people tumble. According to the data of five typical tumble scenes of forward hand use, forward knee use, backward, sideward and empty chair sitting, an extremely abundant training and judgment sample
library is constructed. Omnibearing data coverage enables the
system to deeply understand the slight difference between the normal activity and the tumble event in the dynamic process. Specific
data modeling is carried out on scenes such as stooping, sitting down,
jumping and
lying down which are extremely easy to generate false alarms, and a
system is effectively taught to distinguish'controlled posture change 'and'out-of-control falling process', so that the
false alarm rate is substantially reduced fundamentally, and user trust decline and resource waste caused by frequent false alarms are avoided.