Activity recognition model and system considering universality and individuation

Abstract

The invention relates to an activity recognition system considering universality and individuation. The system comprises a communication framework constructed by a data acquisition terminal, an edge computing device and a cloud computing platform. The activity recognition system performs personnel activity recognition and model updating by means of a communication framework; the edge computing device further comprises a model training module and an activity recognition module. The model training module calls a local activity identification model based on continuous verification of a user ID; and a generalized network structure and a personalized network structure of the local activity identification model are trained by using the first data in a mode of fusing the user personalized features and the model universal features, so that the personnel activity identification process performed by the activity identification module gives consideration to universality and individuation. Besides, the system based on the federated learning framework is combined with the cloud-side communication framework to update the activity recognition model, so that the universality of the model can be improved, and the risk of leakage of personalized features of a user can be avoided.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to an activity recognition model and system that takes into account versatility and individuality. Background technique [0002] In industrial production, health monitoring, security testing, smart home and other scenarios, activity recognition plays a very important role. To recognize user activities, current methods require users to provide a large amount of labeled data to train reliable activity recognition models. However, due to the differences in body characteristics and behavior habits among users, it is difficult for existing recognition models to be directly used to identify activities of new users. Therefore, new users still need to provide a large amount of data to adjust the recognition model parameters or retrain a suitable activity recognition model, which greatly reduces the user experience. [0003] In order to reduce the training cost of new users...

AI Technical Summary

Problems solved by technology

However, in the existing migration methods, the data-based migration method needs to occupy a large amount of network resources, and there is a danger of privacy leakage.

In addition, the method of model migration needs to manually determine the migration part, requires more manual intervention, and the migration cost is relatively high

Images