Modelling human behaviour in surveillance videos is an important research topic due to its applications, such as event summarization, path prediction and abnormal behaviour detection. Furthermore, the development of Ambient Intelligence (AmI) techniques relies on automatic human behaviour understanding, for enabling innovative human-machine interactions and unobtrusive and autonomous decision-making smart environments. The dynamics of motion patterns are relevant at understanding valuable information about human behaviour and can be applied in the analysis of Parkinson’s and Alzheimer’s diseases. Parkinson’s patients experience mobility issues, which are major in the latest stages of the disease, such as slowness of movement and stiffness. In the case of the Alzheimer’s patients, due to the decline of the cognitive abilities, they experience problems related to memory, planning activities or processing of information, which also affect the way they walk. A confusion state can be detected based on motion patterns, which are in this case chaotic or repetitive, as the patients tend to wander around or to scan the room several times due to agitation and nervousness.
Constant sensory monitoring of Parkinson’s and Alzheimer’s patients has many advantages, helping at detecting unwanted or dangerous events, gathering statistics about their daily activities and decreasing the load of the caregivers. The analysis of the sensory information is performed using spatiotemporal information and context information such as semantic regions in a scene, which form the set of features. Next, the extracted features are fed to a classification algorithm for detecting several types of behaviours (e.g. stationary, normal or abnormal behaviour). Training of the algorithms was done on a dataset recorded at Maastricht University, consisting of 50 subjects which were asked to perform a set of activities, including normal activities, such as preparing tea or coffee and activities which elicit confusion, such as searching for an item which is not present in the room. The generalization ability of the machine intelligence algorithms is demonstrated by testing the developed system in new environments, such as the Living Lab in Madrid (Spain) with Parkinson’s patients and in the clinic of Psychiatry and Psychotherapy in Pecs (Hungary) with Alzheimer’s patients.
An example of recorded trajectories is depicted below, where on the left the trajectory of a healthy elderly person can be noticed and on the right the trajectory of a Parkinson’s patient is shown. In the case of the Parkinson’s patient, the movement degradation can be noticed, as he is stopping more often and has difficulties in exploring the space. The motion pattern analysis is useful at supporting the early diagnosis of the disease and in the prevention phase. An explanatory video can be watched on youtube.
Figure 1: Example of trajectories of a healthy elderly (left) and of a Parkinson’s patient (right).
The analysis of Alzheimer’s patients is useful at detecting confusion and repetitive behaviours, while a video can be watched on youtube. In case a confusion state is detected using machine intelligence algorithms, an alarm is triggered, and the caregiver of the patient is notified.
The system developed for analyzing motion patterns and behaviour patterns of Parkinson’s and Alzheimer’s patients showed promising results, supporting the increasing trend of independent living and contributing to improving the life quality of elderly people living with physical or cognitive disabilities. However, more efforts will be invested by researchers and medical professionals to improve the system performance, address its limitations and perform extended tests with patients during the ICT4Life pilots for offering an efficient and sustainable solution.
Department of Data Science and Knowledge Engineering
Picture created by Janoonn028 – Freepik.com