IW - Intelligent Wheelchair

Paper: Intelligent Wheelchair (IW) Interface using Face and Mouth recognition

Written by: Jin Sun Ju, Yunhee Shin, and Eun Yi Kim

Comments: Ross

Summary:
This paper proposed the idea of an intelligent wheelchair that would recognize face and mouth controls from a severely disabled person in order to navigate the wheelchair. The direction that the IW would go would depend on where the users face was inclined, and the starting/stopping of the IW would depend on the users' mouth. There are also 10 range sensors used for the wheelchair so that it can avoid obstacles in whatever environment it is in. The wheelchair has an embedded computer and sensors that allow the user to control the movement of the chair. The system is also able to figure out intentional behavior (looking to the right to move the chair) versus unintentional behavior (looking at something that could be a potential obstacle).

This system recognizes a users movements and converts it into something that the chair can understand so that it can move accordingly. There are three steps in this process:

1. Detector
2. Recognizer
3. Converter

Once the facial recognition is in place, it is then time to figure out what the mouth is doing. The way the mouth is detected is based on two properties:

1. The mouth is located in the lower region of the face
2. The mouth has high contrast when compared with its' surroundings

There are a few mouth 'templates' that are taken to begin the recognition of what the mouth is doing (telling the chair to go or stop). K-means clustering is used to figure this out, and then the Hamming distance is used to figure out what mouth template will correspond to a given candidate that is using the chair.

The converter portion is used in order to give a certain amount of power to the machine based on the users input that the compute is reading off. The board program will control the speed by regulating the amount of voltage that the chair will receive.



The authors wanted to test their solution against solutions that have already been created. They tested it against the joystick wheelchairs that most people are used to seeing, and they also compared it to the headband method of moving a chair which is a little bit similar to their solution. Their first experiment dealt with the commands while the second experiment compared their method with the other two methods. They used a range of people in order to test this which is shown below.


Their results showed that their system could accurately detect the face and mouth of each user, and also figure out how their head was inclined so that the IW could move in that direction. The calculated how long each action would take and also compared that to the other methods. They used recall and precision in order to show their methods performance.

They came to the conclusion that the joystick method was the fastest because it more comfortable for people to use because they did not really want to use their face to move the chair. After training the individuals on how to use their face it became evident that their method was more accurate.

Discussion:
I think this paper was very interesting and it would be really nice to have for those people who are severely disabled and are unable to use the joystick wheelchairs. There has been some future work already being planned for this wheelchair such as adding more sensors to accurately detect obstacles to avoid injury to the user of the chair. I think that is a great idea because there are multiple things that could be seen as an obstacle to a wheelchair and it is necessary for this chair to help the user avoid them. They also want to make sure that IW can better detect situations so that when a user is talking to someone the IW won't start moving around on them. This would definitely be something useful for people and I hope that they can improve on it so that people are more comfortable with using it.