Eyeball gesture controlled automatic wheelchair using deep learning

Traditional wheelchair control is very difficult for people suffering from quadriplegia and are hence, mostly restricted to their beds. Other alternatives include Electroencephalography (EEG) based and Electrooculography (EOG) based automatic wheelchairs which use electrodes to measure neuronal activity in the brain and eye respectively. These are expensive and uncomfortable, and are almost impossible to procure for someone from a backward economy. We present a wheelchair system that can be completely controlled with eye movements and blinks that uses deep convolutional neural networks for classification. We have developed a working prototype based on only a small video camera and a microprocessor that shows upwards of 99% accuracy. We also demonstrate the significant improvement in performance over traditional image processing algorithms for the same. This will allow such patients to be more independent in their day to day lives and significantly improve quality of life at an affordable cost.