If you would like more information about the profile please click
. You will be directed to a new page where you will need
to provide some information about yourself/your business cooperation needs and submit
your expression of interest in the profile . A Network Partner will then contact
you to discuss your interest in greater detail.
A Slovenian research institute is offering an algorithm for activity recognition with a smartphone and an optional wristband. The algorithm uses context-specific machine-learning models and therefore does not depend on the position of the smartphone on the body. The algorithm outperforms several consumer devices in terms of accuracy in real life setting. The algorithm is available via licencing or service agreement to companies and researchers developing wellbeing applications.
Accurate activity monitoring is required in domains where further reasoning or person-specific recommendations rely on the user's physical activity. These range from lighter topics such as sports and lifestyle to sensitive topics such as health. Current market offers smartphone applications and dedicated devices, whose scope is limited to step counters (pedometers) with inaccurate estimation of energy expenditure or to monitoring only certain type of activities that make them unsuitable for applications that require accuracy.
There are many activity monitoring applications and devices on the market which mostly rely on rather simple techniques such as step counters from which they estimate the energy expenditure or they require the user to explicitly state which activity is being performed (e.g. they do not perform any automated activity recognition). The evaluation of consumer devices has shown that they also lack accuracy in real-life every day activities. Algorithms that are more accurate have been developed in the domain of body sensor network where the user is required to follow the instructions about the exact placement of each of the used dedicated sensors which makes the technology hard to use.
The artificial intelligence researchers from a Slovenian public research institute have developed, evaluated and validated an algorithm for real-time continuous activity monitoring that utilizes sensor data from a smartphone or wrist-worn or chest-worn device and can fuse the data and decisions of both devices if both are present on the body.
The activity recognition and estimation of energy expenditure is done using machine-learning techniques. The design of the algorithm enables the user to put the smartphone in any pocket or in a bag and in any orientation since the algorithm first detects the presence of the devices and then normalises the orientation. The normalised signal is used to recognise the location in case of the smartphone. The recognised context of the sensors is used to select appropriate context-specific machine-learning model for activity recognition and estimation of energy expenditure.
The technology is available either:
a) under the services agreement as a Software-as-a-Service (SaaS) through application programming interface (API) or
b) under licencing agreement for the algorithm which can be run on a smartphone.
Authors of the algorithm are computer science experts employed at the Slovenian institution for research in sciences and technology. They are specialized in development of proprietary methods and algorithms for analysing wearable sensor data used mainly in health domain, but applicable to other domains. The team has been among finalists of the global competition for medical diagnostic devices. They have won the international competition for activity recognition. They are active in several projects for development smart watch monitors for independent living of seniors with dementia; detection of falls and abnormal behaviours for elderly; support older workers in reducing physical and mental stress using wristband and personalized advices and decision support to help patients with heart problems.
The researchers are looking for partners which are either:
a) companies or research institutions which develop applications and would need the service (SaaS) through API (service agreement) or
b) companies which are interested in obtaining a licence for implementation of the activity monitoring algorithm in their application (licencing agreement).
In particular following companies or research institutions active in wellness and health domains are sought:
- companies that develop and produce wearable wireless wellbeing, sport and fitness devices;
- companies that offer solutions for remote patient or elderly monitoring, on-site professional healthcare monitoring and home/office/work environment monitoring.
Advantages of the real-time continuous activity monitoring algorithm:
- It self-detects the presence of the devices
- It can monitor the activities with smartphone, wristband or chest strap alone or with a combination of two devices (smartphone + wristband, smartphone + chest strap)
- The devices can be worn in any orientation since the algorithm detects it and normalises it
- The smartphone can be worn on three locations (trousers pockets, jacket pockets, bag) which are recognised automatically utilising machine-learning techniques
- The activity-recognition models are trained on the most typical every day activities (rest, home chores, gardening, eating, etc.) as well as sports activities (walking, Nordic walking, running, cycling, etc.).
- The models for estimation of energy expenditure are trained on the data labelled by the indirect calorimeter.
- The evaluation results show that the algorithm outperforms several consumer devices.
The researchers evaluated the algorithm in the laboratory environment where they compared it against cold standard indirect calorimeter and against two commercial devices (Microsoft Band 2 and Bodymedia Fit). The results show that the system is able to recognise the activities with the 85% accuracy on average with a single device and 92% on average when two devices are present. The evaluation of energy expenditure shows error of 0.6 MET (Metabolic Equivalent of Task) on average. The researchers compared results against commercial devices which they outperformed in terms of MET and in terms of kilocalories. They validated system in two pilots (diabetes patients – 3 months, children – 1 month). The researchers are currently in preparation for the 6 months pilot with aging workers.
Complete solution is available for demonstration. The algorithm is being constantly improved and tested. The authors of the algorithm have necessary knowledge to offer all the engineering expertise and custom development support (R&D team of 30 people) to the potential partner to implement the algorithm in a commercial product.
The research institute is looking for:
a) industrial partners who are interested in obtaining a licence for activity monitoring algorithm – licencing agreement and
b) companies or research institutions who would like to use the service (SaaS) through API - service agreement.
These are especially:
- companies which develop and sell health care and lifestyle applications;
- companies which develop, produce and sell wearable wireless wellbeing, sport and fitness devices;
- companies which offer solutions for remote patient or elderly monitoring, on-site professional healthcare monitoring and home/office/work environment monitoring;
research institutions active in ambient intelligence, wellbeing, e-health, elderly support, sport tracking and other activity related research and development activities.
Partners interested in service agreement will be able to use the activity algorithm in their application via SaaS service, which is offered by Slovenian research institute.
Partners interested in obtaining a licence for the activity algorithm are expected to be able to integrate the source code of the activity algorithm in their applications, e.g. on a smartphone, wrist-worn or any other smart wearable device. The authors of the activity algorithm are able to custom develop the source code in case of specific requirements of a partner.