In recent years, human activity recognition (HAR) has attracted a lot of attention due to its wide application, such as indoor positioning and navigation. This paper proposes a MconvLSTM to construct a multi-unit deep network structure,which can effectively improve the accuracy of HAR. Firstly, the input data is dimensionally expanded. Secondly, multiple convLSTM module are used to input data from different sensors to achieve partial weight sharing. Multiple outputs are merged finally. The experimental results show that the partial weight sharing mechanism and dimension expansion effectively improve the extraction of single sensor features, aiming to improve the activity recognition rate. Using public UCI datasets for testing, the accuracy is significantly improved compared to traditional convLSTM network results.
Paper
Full text
Multi-convLSTM neural network for sensor-based human activity recognition
Semantic Scholar · Computer Science · 2020
Abstract
In recent years, human activity recognition (HAR) has attracted a lot of attention due to its wide application, such as indoor positioning and navigation. This paper proposes a MconvLSTM to construct a multi-unit deep network structure,which can effectively improve the accuracy of HAR. Firstly, the input data is dimensionally expanded. Secondly, multiple convLSTM module are used to input data from different sensors to achieve partial weight sharing. Multiple outputs are merged finally. The experimental results show that the partial weight sharing mechanism and dimension expansion effectively improve the extraction of single sensor features, aiming to improve the activity recognition rate. Using public UCI datasets for testing, the accuracy is significantly improved compared to traditional convLSTM network results.