Interaction-based Human Motion Analysis

In this project, we propose a new method to analyze human motion based on the interaction with the surrounding environment, which provides a better understanding about the nature of the performed motion and enhances the performance of modern motion-related applications.

Understanding human movement is a central problem for motion-related applications such as behaviour monitoring for smart homes and movement evaluation for physical therapy. Most traditional 3D motion analysis algorithms are human-centered, meaning that they only consider features of the human body but not the interaction with the surrounding environment. Imagine the movement of an older person standing on the floor while doing some arm-stretching exercises, and another standing on a chair to fix a light bulb, which is considered to be dangerous. The two motions have completely different contextual meanings, but are surprisingly similar in terms of human body posture. Traditional computer-based motion analysis methods disregard the relationship between the human and the environment, and thus cannot accurately tell the difference between the two motions.

We observe that real humans usually comprehend the context of a motion based on its interaction with the surrounding environment, such as sitting on a sofa, watching the television, riding a bicycle, etc. We believe that by considering these types of interactions, higher quality motion analysis can be improved as a result of better understanding on the context of the motion. Therefore, we propose a new algorithm to analyze human motion based on the interaction between the human and the surrounding environment, which will enhance the accuracy of motion identification and the performance of movement evaluation. Our algorithm evaluates detailed 3D movement features with respect to the environment, such as analyzing the subtle movement of the feet of a Parkinson's disease patient with respect to the position of the stairs during a stair climbing motion. It can therefore (1) analyze movement features from the interaction point of view, and (2) identify what kind of motion the user is performing based on the interaction context.

The system proposed in this research could be used to enhance the quality of life of older people to enable greater independence and reduce the burden on emergency and care services caused by a rising ageing population. Our algorithm accurately identifies human motion, which is an important step towards a smart home system that takes care of older people autonomously. It also aims to evaluate human movement, which can significantly reduce the labour cost of rehabilitation and coaching for older people, as well as early stage motion-related disease (such as the Parkinson's disease) diagnosis.

This research aims to enhance the quality of life and reduce the cost of supporting older people, which are major challenges in the UK due to population ageing. This target will be achieved via a new algorithm that better identifies and analyzes human motion, such that services that are usually required by older people including smart home, rehabilitation/sport coaching and movement disorder diagnosis can be enhanced/automated.

According to the 2011 census, the population of the UK aged 65 and over was 10.4 million, which is equivalent to 16 per cent of the UK population. The Office for National Statistics has projected that in England in 2030, there will be 51% more people aged 65 and over compared to 2010. By 2050, a 65-year-old man in Britain can expect to live to 91, whereas in 1950 his life expectancy was 76. This trend presents significant challenges for services we depend on to take care of older people and also for an increasing proportion of the working age population which provides informal care. This research directly addresses this issue by proposing a new technology for motion analysis, which is a key part of several major services required by older people, including smart homes, rehabilitation and movement disorder diagnosis.

Smart home, also known as home automation, describes technologies that facilitate in-house support such as automatically calling for help when accidents occur. An ideal smart home allows older people to live in their own home while maintaining access to high quality support, minimizing their dependency on costly healthcare facilities like care homes. One major challenge of smart homes is to understand what the user is doing, in order to decide how to support the user. The proposed research can accurately identify dangerous behaviours like standing on a chair to fix a light bulb, as well as accidents such as falling and colliding with obstacles. With this kind of technology, the labour cost of supporting older people can be reduced, and older people can live with greater independence.

Rehabilitation and sport coaching are two major services used by older people because of the deterioration of physical fitness. The former focuses on recovering from physical injury, while the latter is to strengthen body fitness and prevent injury. The current main streams of these services are highly labour intensive, involving professional trainers to guide the user throughout the training process. Due to high running costs, the availability of these services is under heavy stress, and will likely worsen as a higher proportion of the population lives longer. This research can automatically evaluate the motion of a person when he/she interacts with the environment. This is especially useful for rehabilitation and sport coaching, in which the user usually needs to interact with equipment such as riding exercise bikes and stepping over obstacles. The proposed method can analyze how well the user handles the equipment, and point out how body parts should move with respect to the equipment. It is an important step towards autonomous rehabilitation and coaching.

Movement disorder diagnosis is another service often required by older people. This is because many health issues suffered by older people such as Parkinson's disease, muscle injury, and degenerative joint disease, result in movement disorder. Early stage diagnosis of these issues usually involves physiologists observing the degree and the symptoms of the disorder. However, even for professionals, it is a challenging problem to accurately diagnose these diseases. The proposed algorithm can evaluate information considering the relationship between the motion and the environment, such as the subtle movement of the feet with respect to the stair in a stair climbing motion. Such kinds of human-environment relationship features have not been fully explored in traditional research, and can potentially improve diagnosis accuracy.