| Human Activity Recognition（HAR）is an important research content in the field of human-computer interaction,and has a wide range of applications in the fields of virtual reality,intelligent interaction,and smart medical care.Activity recognition can identify a variety of individual activities,such as running,walking,etc.,and has a wide range of applications in context-related fields.HAR mainly uses computer equipment to collect individual activity data and model the data to identify individual activity.At present,sensor-based activity recognition models can be roughly divided into two categories: traditional classification models and deep learning models.The current models that perform well in the field of activity recognition are often constructed using deep neural network technology.Compared with the way of manually designing features for data processing,models built using deep neural network technology can often automatically extract features without manual intervention.Models based on deep learning technology have great advantages over traditional models in terms of cost and difficulty.Activity recognition models based on neural network technology often use convolutional networks for feature extraction in the spatial dimension,and then use recurrent neural play for feature extraction in the time dimension.Not only that,some models also use the Attention mechanism to further enhance the generalization ability of the model.There are two major deficiencies in the current datasets used to train activity recognition models: First,sensor data and activity labels are mostly collected using special experimental equipment in a supervised experimental environment.Affect the normal life of users;Second,the collected sensor data is less of a variety,and the data modalities are relatively single.As the manufacturing process of smartphones has advanced,today’s smartphones tend to be equipped with a large number of physical sensors,making them excellent tools for data collection.Aiming at the two deficiencies of existing datasets,we builds an experimental platform for activity recognition data collection and activity label collection: iSense.iSense can automatically collect a large amount of sensor data through a small amount of user operations in the user’s daily life.This data collection method takes full advantage of the powerful functions and widespread availability of smartphones,reduces the cost of data collection,and can collect richer data from a larger population.We builds our own activity recognition model Attn Fusion,and uses the sensor data and labels collected by the iSense experimental platform for user experiments to train the model.This paper compares the training results of the Attn Fusion model with several other representative models,and verifies the advantages of the Attn Fusion model compared to existing models in activity recognition.After the model training is completed,this paper integrates the model with the iSense client to realize real-time activity recognition on Android smartphones. |
