Thursday 06 March 2025
A new system has been developed that can recognize human activities in real-time using skeleton data and deep learning techniques. The system, which uses a combination of EfficientNet and ConvLSTM, is capable of detecting a wide range of activities, from simple movements like walking or sitting to more complex actions like sneezing or falling.
The technology has the potential to revolutionize healthcare by enabling doctors and caregivers to monitor patients’ health in real-time. For example, it could be used to detect falls in elderly patients who are at risk of injury or to identify early signs of neurological disorders such as Parkinson’s disease.
To develop the system, researchers collected a large dataset of skeleton data from various human activities. They then trained their model using this data and tested its performance on two publicly available datasets, NTU RGB+D 120 and HMDB51.
The results were impressive, with the system achieving accuracy rates of over 94% for cross-subject evaluations and over 96% for cross-view evaluations on the NTU RGB+D 120 dataset. The system was also able to detect activities in real-time using a Raspberry Pi and GSM module, delivering alerts via SMS to caregivers or patients.
The researchers hope that their technology will be used to develop wearable devices that can monitor people’s health in real-time. This could have significant benefits for patients with chronic conditions, allowing them to receive timely interventions and preventing complications.
The system is not without its limitations, however. For example, it may struggle to detect activities in environments with poor lighting or background noise. Additionally, the dataset used to train the model was limited to a specific set of activities, which may not be representative of all possible human actions.
Despite these challenges, the researchers are optimistic about the potential of their technology. They believe that it could have far-reaching implications for healthcare and quality of life, and they look forward to continuing to develop and refine their system in the future.
Cite this article: “Real-Time Human Activity Recognition System Using Skeleton Data and Deep Learning Techniques”, The Science Archive, 2025.
Human Activities, Deep Learning, Skeleton Data, Efficientnet, Convlstm, Healthcare, Real-Time Monitoring, Wearable Devices, Chronic Conditions, Patient Care
