The goal here is to use our camera in the back to detect when a stair is finished. We would like to create a small neural net that can be taught to recognize what a stair is an what not. There are many special cases when reaching the top and there could be lying objects around. The algorithm still has to detect with 99.999% certainty that the stairs are finished. On this technology we could then build up to start detecting steps when driving in balancing mode. This will be much more difficult and does not have to be scope of the thesis.
Wheelchair drivers are confronted with unique challenges in their daily life, due to their limited mobility. In older buildings and in cities with insufficient infrastructures stairs are one of the biggest obstacles they might encounter. Therefore the goal in the Focus Project called Scewo developed an electrically powered wheelchair that will climb stairs and curbs without external help. The wheelchair balances and drives on flat ground using two wheels but can also climb stairs using two tilting tracks that lower when needed. Additionally, there are two retractable wheels located at the back of the chair that act as a support system. These wheels help with the transition at the top of the stairs and allow the chair to be parked at a spot. The project was supervised by the Autonomous Systems Lab (ASL) at ETH Zurich. When showing our prototype to the public the feedback towards the device was so positive and a part of the team decided to found a company to bring the product to the market. Scewo is now an ETH-Spinoff since 1.5 years we are close to delivering our first units to the customer.
- Research different methods to train a system to detect steps
- Evaluate a processing system to run the code
- Extract stair distance, stair angle and other relevant information from a stream of images
- Implement a basic version of the algorithm on the system
- Test the algorithm with the Scewo wheelchair