UltaNet: An Antithesis Neural Network for Recognizing Human Activity Using Inertial Sensors Signals

Human activity recognition (HAR) is an essential component of ambient assistive living. HAR has traditionally relied on computer vision techniques. However, it has several drawbacks, including lack of privacy, higher operational costs, and being constrained by the number of spaces available for cameras, so it cannot be used for applications that require long-term monitoring of people. The use of initial sensors is proving to be a vital solution for HAR. Smartphones and smartwatches have embedded accelerometers and gyroscope sensors that help deep neural networks for reliable activity recognition and classification problems. In this letter, we have presented a novel deep neural network model, which is opposite to the traditional models. It has a gated recurrent unit layer followed by different kernel-sized convolutional layers. Our model has been evaluated on an openly available dataset by the wireless sensor data mining lab, and comparison with previously existing models proves that it outperforms them.