Research On Human Activity Recognition Methods Based On Wearable Sensors

Human Activity Recognition(HAR)is not only widely used in military,anti-terrorism,national security and other fields,but also in daily life such as medical monitoring and health monitoring.HAR has greatly improved the level of rehabilitation care for patients and the quality of disabled persons’ life,and reduced the probability of accidental injuries for old people.Recently,there are many technologies used in the HAR field,such as visual recognition,signal recognition,and sensor recognition.Although visual recognition has good recognition accuracy,it has the disadvantages of privacy leakage,easy occlusion,and light sensitivity.While signal recognition has the advantages of device-free and privacy protection,it is,sensitive to environmental noise and is not suitable for crowded environment.So sensor-based human activity recognition is a better choice.However,current human activity recognition methods based on wearable sensors have problems such as low recognition accuracy and poor fault tolerance.Therefore,this paper proposes a multi-level decision recognition method based on multi-sensor fusion.Besides,the paper improves accuracy through deployment optimization.First,the paper proposes a multi-level decision recognition method based on multi-sensor fusion aiming at low accuracy and fault tolerance.The method improves the accuracy of activity recognition through multi-sensor decision-level fusion and a multi-level decision model.At the same time,the model has low coupling between sensors.Therefore,high fault tolerance can be ensured when the sensor fails and the entire activity recognition system has higher fault tolerance and availability.Second,different sensor deployment schemes will have different recognition accuracy in wearable sensor-based activity recognition.Traditional empirical deployment solutions cannot guarantee optimal sensor deployment.In order to improve the recognition accuracy and consider the cost of sensor deployment in practical applications,the paper proposes a CM-WOA-based multi-sensor deployment optimization method.This method balances the recognition accuracy and the cost of sensor deployment.The method not only improves the recognition accuracy,but also reduces the number of sensors.Finally,the effectiveness of multi-level decision recognition method based onmulti-sensor fusion and the CM-WOA-based multi-sensor deployment optimization method is verified by experiments.Besides,the paper simulates the sensor failure situation in the real scene,and verifies that the proposed activity recognition method has high fault tolerance.