The invention discloses an embedded intelligent detection method for a human tumbling state of a wearable device. A six-axis acceleration sensor is placed at a human wrist, original three-dimensionalacceleration data of a human body in the three-dimensional direction are extracted, sent to an ARM embedded CPU and processed. The ARM embedded CPU processes the collected original three-dimensional acceleration data and judges whether the posture is a tumbling posture; the original three-dimensional acceleration data are preprocessed to obtain simplified three-dimensional acceleration data; the simplified three-dimensional acceleration data is subjected to characteristic extraction, the characteristic threshold range, continuous posture recognition threshold and tumbling threshold are set todetermine whether tumbling occurs. The embedded intelligent detection method for the human tumbling state of the wearable device pre-processes the data, calculates and judges the characteristic data,classifies and refines the types of posture, better improves the recognition of tumbling and postures, and makes the recognition more accurate and reasonable.