Human activity recognition using residual deep network with extreme learning
Garima Bohra, Chandra Kumar Jha, Neelam Sharma · Computational Methods in Science and Technology · 2024
In recent years, human activity recognition (HAR) has emerged as an important topic of research due to the multiple applications it has found in various fields, including fitness and strength surveillance, residential care, biometric verification, aged care, and many others. Sensors are a great alternative for activity recognition because of their widespread use. The bulk of wearable sensors, including accelerometer, gyroscope, camera, GPS, microphone, compass, and others, are built into the newest smart devices. Such sensors measure different properties of an object, are simple to use, and are inexpensive. Researchers using machine learning (ML) now have new potential to accurately recognize human behaviours thanks to the usage of sensors in the HAR sector. Deep learning (DL) is gaining popularity among HAR researchers because it outperforms standard machine learning (ML) approaches. In this paper, hybrid features of mechanisms and learning are used by employing deep learning and machine learning; deep learning uses RESNET-50 and machine learning uses XG boosting. In experiments, the proposed approach was validated by a 10-cross comparison with the state-of-the-art approach. In experiment also classified human activities and intruders.