This final year project aims to implement edge detection and saliency detection to down sample a high-resolution video feed into vital information that can be conveyed with a cortical visual prosthesis. This must all be accomplished using a relatively affordable, small portable single board computer to allow broader adoption of the technology.

To achieve the requirements of the project, a Raspberry Pi 4 and a Raspberry Pi Camera V2 was chosen as the single board computer of choice. Developing on the Raspbian OS on the Raspberry Pi, the programming language of choice was Python 3, due to the large number of supporting libraries and compatibility with Raspberry Pi.

As this project aimed to determine the feasibility of using a single board computer to achieve the image processing requirements of the cortical visual prosthesis, large amounts of resources were dedicated to experimenting and finding the optimal solution with limited computing resources. From TensorFlow and other deep learning libraries to OpenCV, many options were tested to determine whether the project requirements of pseudo real-time image processing can be achieved on the Raspberry Pi.

Using the Raspberry Pi Camera V2 input, the camera feed is processed using Python using a Canny edge detector and Fine-Grained saliency detector, coupled with multiple layers of Gaussian blur and Thresholding to produce a video feed of the most important edges. This feed can then be further down-sampled and transmitted via GPIO to a cortical visual prosthesis.

The base requirements of this project has been met satisfactorily, and the processed video feed allows users to navigate their immediate surroundings with sufficient ease. To achieve higher accuracy using neural networks, a more powerful processing medium should be considered, though the current iteration of this project is sufficient for the basic requirements for the cortical visual implant.