The disclosure is related to a method and an apparatus for human fall detection with power-saving feature. In the method, a processor is set in a sleep state under a normal operating condition, and sensor data generated by a sensor unit of the apparatus worn on a person is stored in a buffer. A processor of the apparatus is woken up from a sleep state when the processor receives a collision signal, and the processor then retrieves the buffered sensor data and current sensor data. The sensor data is processed by a fall detection program. The apparatus will generate an alarm for a fall event if the sensor data meets the preset fall conditions； otherwise, the processor returns to the sleep state. The apparatus saves power by avoiding huge data calculation while staying in the sleep state until being woken up by the collision signal.
