Extraction of 3D image information based on iso-disparity concepts for a visual prosthesis
Benyamin Kheradvar1 *, Amir M. Sodagar2 , Amir Mousavinia3
- Faculty of Electrical Engineering, K. N. Toosi University of Technology
- K. N. Toosi University of Technology and York University
- Faculty of Computer Engineering, K. N. Toosi University of Technology
Abstract: According to the World Health Organization’s 2019 report on vision, 2.2 billion people are blind or visually impaired. In recent decades, various types of portable, wearable, and implantable devices have been introduced for the navigation of blind subjects, the restoration of their sight, or providing them with artificial vision. From among such devices, visual prostheses are implantable microsystems that restore vision through receiving a live stream of externally-captured image information, and stimulating the retina accordingly. This work presents the results of an innovative ground plane detection approach, which aims at effective reduction of the image data transferred to a visual prosthesis yet preserving the key image information.
Methods: Previously, the authors proposed a novel approach for automatic ground plane detection in artificial vision systems. That approach was dedicated to disparity-map images provided by stereo vision techniques, and took advantage of the patterns existed in the Iso-Disparity Strip Width (IDSW) associated with ground planes. In the present work, the aforementioned ground plane detection algorithm is adopted and implemented on a hardware/software platform for use in a retinal visual prosthesis. To prepare the detected ground plane, it is downsampled to the size of the electrode array of the system in two steps: First, the ground plane is downsampled based on the statistical ‘mode’ of 2D zones of the image pixels. Then, the zones labeled as ‘ground’ are aggregated with the neighboring zones that comprise more than 25 percent of ground pixels. To visualize the perception of an implantee from the stimulations generated on the electrode array, the resulting ground plane is then phosphenized using a 2D Gaussian kernel.
Results: Two datasets of synthesized and captured stereo images as well as a synthesized stereo video are used to demonstrate and qualitatively evaluate the results. The captured dataset is prepared by the PROSPECT (PROsthetic Sight Performance Evaluation and Code Test) setup, which is a simulated prosthetic vision (SPV) developed by the authors.
Conclusion: This work reports on preparing the ground plane detected in a live stream of captured images in order to be transferred to a visual prosthesis, and also simulates the image precepted by an implantee.
