Activity detection examples with multiple sensors
depth sensor
|
The depth sensor has the advantage of working during the night. This is important to detect activities in low visibility situations. Such as falls near the bed area. The video contains one of those situations. The main drawback is that the sensor does not perform very well in situations of strong natural light. For example, when the person is near a window.
A positive property of the sensor is its precision, it is possible to detect people and their joints with high accuracy. Nevertheless, the detections are confined to a distance range of 7m approx. |
Color camera
|
Color cameras have the advantage of working in long distance ranges. The accuracy of detection of pedestrians is strongly related to the environment configuration and the configuration of the algorithm. These sensors normally cannot work during the night. When they operate in night mode, the visual signal is very poor making hard to detect and track meaningful fine grained activities.
For more details about the algorithm read: Unsupervised Discovery of Human Activities from Long-Videos Discovery of human activities in video |
|
Thermal sensor
Thermal sensors open new possibilities for activity recognition. "Seeing the heat" can be used to detect hazards -like forgetting the stove on-. Tracking pedestrians is less complicated since the signal noise ratio is better than with other "visual" sensors. Also, this sensors can work 24/7 with meaningless loss of signal, while the other "visual" sensors get into trouble with strong light conditions or at night. We are yet building machine learning algorithms to take the full advantage of thermal signal with the goal of achieving automatic semantic interpretations.
|
|