In this paper a novel scalable model is presented to perform indoor human activity recognition, assigning an activity to every recorded time step. This yields a vector of values containing speeds of objects moving to and from the radar. Using the superpositions of the reflections from different moving parts of a human body, a Micro-Doppler (MD) signature 4 can be generated. The reflected signal arrives back at the radar system after a certain amount of time from which the distance, angle of the object in relation to the sensor and speed can be derived. This phenomenon is known as the Doppler effect. If these objects are moving, the frequency of the signal shifts. Radar systems transmit electromagnetic radio signals that will be reflected by objects. Finally, in low light conditions, these systems also outperform traditional video-based surveillance systems. Secondly, the signals can penetrate walls and see through other materials 3, allowing for monitoring in places where regular video equipment would not be installed for surveillance (e.g., a bathroom). Firstly, they are privacy preserving as no directly interpretable image can be extracted from the radar data. This is where compact and cost-effective radar devices can alleviate some issues. Another possibility is to provide the patient with a wearable device that can detect when a fall occurs, but this requires the patient to wear this device at all time, which may be forgotten or lost. To combat this, staff could set up cost-effective video cameras or other surveillance methods, however this is too privacy invasive. This is most meaningful for geriatric patients 1, 2 that are more likely to suffer lasting injuries from a fall, especially if treatment is delayed. Real-time activity recognition in a hospital room or nursing home is important, because it can help to detect troublesome events, such as the fall of a patient, as soon as possible. This enables fast notifications to staff if troublesome activities occur (such as falling) by the on-premise device, while the off-premise device captures activities missed or misclassified by the on-premise device. Next, a part of the calculation and the prediction is sent to a more capable off-premise machine (most likely in the cloud or a data center) where a backward RNN calculation is performed that improves the previous prediction sent by the on-premise device. First, a forward Recurrent Neural Network (RNN) calculation is performed on an on-premise device (usually close to the radar sensor) which already gives a prediction of what activity is performed, and can be used for time-sensitive use-cases. This work presents a framework that splits the processing of data in two parts. Often these networks are large and do not scale well when processing a large amount of radar streams at once, for example when monitoring multiple rooms in a hospital. This can be achieved by using Deep Neural Networks, which are able to effectively process the complex radar data. Radar systems can be used to perform human activity recognition in a privacy preserving manner.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |