HAR-HEALTH: Recognition of Human Activities Associated with Chronic Diseases
The project addressed the research and development of new algorithms for human activity recognition, based on visual information and, mainly, sensor data from mobile devices. Unlike the state of the art, it was desirable to recognize groups of activities being performed simultaneously by a particular person, as well as to create a profile of this individual by storing their activities over time. To do this, several research topics were addressed with an innovative approach: object detection, tracking and re-identification of individuals, human-object interaction modeling, activity recognition, algorithms for processing and interpreting data from mobile device’s sensors, merging and normalization of signals, among others.
The development took place in close partnership with a group of researchers from Samsung Electronics at the Amazon region, whose technical skills and scientific knowledge have strong synergy with the university’s resources. Therefore, both teams worked step by step on the various strategies created throughout the project in order to achieve the best accuracy rate in the recognition of human activities.
The project also had collaboration from the C.E.S.A.R Institute, which initially provided an application on the Android and Tizen platforms for the collection of sensory data that was used in the development of the project. Later, C.E.S.A.R. received, from this project, classification models based on neural networks and SVM. These models are used in the human activity recognition application, which was developed by this entity.
Contribuições geradas à partir dessa pesquisa…
Object-based Temporal Segment Relational Network for Activity Recognition Inproceedings
In: Conference on Graphic, Patterns and Images (SIBGRAPI), pp. 1-8, 2018.
In: 26th European Signal Processing Conference (EUSIPCO 2018), pp. 1-5, 2018.
In: 12 (7), pp. 1387–1394, 2018.
In: IEEE International Conference on Advanced Video and Signal-Based Surveillance (AVSS), 2016.