Very pleased to learn that our system for Semantic Classification of Human Activities was included in the demo sessions presented during the opening of the new NTT DATA office in Shoreditch, London (link). This is the kind of accomplishment I wanted to follow up with after I introduced the concept at the end of last year.
Kudos to the NTT DATA Romania development team that made this thing possible!
Our system can be seen in the presentation video at min 1:11 - 1:15. Combining control functionalities (implemented as an AI trained 3D gesture recognition algorithm) with a gaming experience (the user had to use marshaling hand signals in order to direct an aircraft to its parking stand) the demo was designed to convey in an interactive / engaging way the type of potential real-life use cases.
Besides the type of activities required by the demo, the system was trained to recognize as well more general full-body type of activities like walking, running, jumping and also specific activities like talking on the phone.
Due to its modular nature, the system enables further developments in which the classifier could be trained to recognize new sets of activities.
The recognition of human activities is becoming more and more important with applicability in different fields like health and well-being applications, human-machine interaction, education, training as well as crowd analysis for urban planning and public safety.
To this end, our system for Semantic Classification of Human Activities represents an AI platform that can be employed for training algorithms to recognize various activities in a way that connects the structural with the functional classification.
The advantage of having such a generalized way of classification is that it is possible to introduce semantic knowledge into the recognition. As such our system allows for a natural way of reporting the recognized activities. Typical example of activities are: Walking around, Opening a door, Talking to someone, Sitting on a chair.
This type of knowledge can be used in situation awareness / intentionality estimation and it can provide the basis for further refinement of the recognition by using of several other algorithms.
留言