A smart walking frame for rebuilding confidence at home
© Anna Lurye - Fotolia.com
Problems with falling are the leading cause of hospitalization among seniors. After returning home, the elderly often limit their movements, lacking enough confidence in themselves to resume normal activities. The European MOBOT project is aimed at assisting them once back home by offering a smart walking frame that adapts to their behaviour. Iasonas Kokkinos from the Galen team is contributing his expertise in image analysis for interpreting real-time videos as part of this project.
What is the goal of the European MOBOT project?
The MOBOT project teams are tackling a complicated problem: how can we help seniors move about safely at home after an initial fall? The solution we will be working on over the next three years is an intelligent support tool for moving about; not the passive walking frame that we are used to but rather, a sort of daily activity companion . Several issues must be faced to develop this new, improved tool. The German team is handling biomechanics, robotics, and haptic data processing, meaning data from pressure sensors on the robotic support platform. The researchers in Greece will be working on analyzing the input acquired by sound and image sensors, such as a call for help. Finally, our own team will be working on image analysis to estimate poses (meaning current position and movement) to detect when the person is, for example, unstable, in order to adapt and provide corrective support.
We will be defining about ten actions so that the tool can understand what that person is doing. This means the tool will be adaptable , ideally even proactive, since it needs to instantly understand that someone wants to get up or sit down in order to hold him or her steady in a position that will provide support, rising on one side to compensate if he or she is losing balance, or even in the more distant future, helping to lift up people who have fallen down.
The tool will ideally be proactive in order to instantly understand that the person wants to get up, and might be able to help them keep their balance
What technical issues remain?
For the visual analysis part, seniors would be watched by cameras from that robotic platform, which would be collecting information about colour, motion, and depth. For depth information, we would need to use a Kinect©-style camera, which means adjusting the software to fit the contours of the problem . This means we need to be very accurate in defining motions.
To do so, we are combining databases of motions made by healthy people, and knowledge of the dynamics of balance, so that the person's next position can be deduced at every moment. This combination gives us the chance to refine the model to detect any changes in balance or any anomalies. The hard part will be selecting which details to analyze : Analyzing an image involves looking at the previous and next frames to understand the motion as a whole and not issue false alarms. At the same time, it is necessary to not consider a stray move of the hand to be "noise" if it could mean the person wants to do something else.
Analyzing the image using an algorithm to detect body parts (this example is for the arms alone)
What about algorithm issues?
We will use the progress the team has made to this point and adapt that learning to the specific challenge of this senior assistance tool. First, we will work on the statistical learning technique that requires extracting descriptions of depth, colour, intensity, and motion for each point in an image, and combining them all into an overarching decision. The team was already quite far along in this work, but adding in the factor of depth means that the algorithms have to be redone. These characteristics will make it possible to deduce an estimated optimal pose .
Next, another matter to work on is efficient object detection . In this project, it is necessary to separate out the modelling for different body parts in order to treat them with greater accuracy, but then the problem becomes how to put them all back together into a coherent whole.
The project is just beginning, so there are many things that need to be put in place. We need motivated, passionate doctoral students to make progress on this project, which is both helpful to society and quite exciting.
These articles could interest you:
- MOBOT: Intelligent Active MObility Assistance RoBOT integrating Multimodal Sensory Processing, Proactive Autonomy and Adaptive Interaction
- Three-year European project
- Four academic partners and three in the private sector (two clinics that will conduct experiments with seniors, and one robotics partner)
- Countries involved: Germany, Poland, Greece, and France. Coordinator in Munich.
- MOBOT project website
- Personal page of Iasonas Kokkinos
Iasonas Kokkinos , research officer in the Galen team