Enhancing scene structure in prosthetic vision using iso-disparity contour perturbance maps

Area of Science:

• Computer Vision
• Robotics
• Assistive Technology

Background:

• Existing methods for prosthetic vision rely on visually salient features or surface fitting, which can be computationally expensive and may miss crucial details like trip hazards.
• Detecting structurally significant features is vital for safe navigation and mobility, especially for individuals using prosthetic vision.

Purpose of the Study:

• To develop a novel, real-time approach for enhancing structurally significant scene features to improve safe mobility for users with prosthetic vision.
• To overcome limitations of existing methods by detecting subtle yet important features relevant to navigation.

Main Methods:

• A new feature based on iso-disparity contours derived from dense disparity images is proposed.
• A cost function comparing local iso-disparity contour orientations is used to detect regions of significant structural change.
• The approach emphasizes features like surface boundaries and clutter in the visual output.

Main Results:

• The novel method effectively detects and enhances structurally significant features, including small, low-contrast obstacles.
• The approach operates in real-time without requiring computationally intensive surface fitting.
• Experimental results demonstrate both quantitative and qualitative validation of the proposed method's effectiveness.

Conclusions:

• The iso-disparity contour approach offers a computationally efficient and effective solution for enhancing scene understanding in prosthetic vision.
• This method has the potential to significantly improve the safety and independence of individuals with visual impairments.
• The technique reliably extracts and emphasizes features critical for safe navigation and mobility.