Abstract
Purpose: Efficacy of current visual prostheses for object recognition is limited. Various limitations such as low resolution and low dynamic range need to be addressed, but here we focus on reducing the impact of background clutter on object recognition. We have proposed the use of motion parallax via lateral scanning with head-mounted camera and computational stabilization of the object of interest (OI) to support background decluttering. We used head-mounted display (HMD) simulations to mimic the proposed effect and test object recognition in normally-sighted subjects.
Methods: Images (24° field of view) were captured at multiple viewpoints, always centered on the OI, and presented in low resolution (20×20). Experimental conditions (2×3) included: clutter (with or without) × head scanning (single viewpoint, nine coherent viewpoints corresponding to subjects’ head positions, and nine randomly presented viewpoints). Subjects utilized lateral head movements to view OIs on the HMD. Each object was displayed only once for each subject.
Results: The median recognition rate without clutter was 40% for all head scanning conditions. With clutter, performance dropped to 10% in the static condition, but was improved to 20% in the coherent and random head scanning conditions (corrected p = 0.005 and p = 0.049, respectively).
Conclusions: Background decluttering using motion parallax cues and not the coherent multiple views of the OI improved object recognition in low-resolution images. This method did not fully eliminate background impact on recognition.
Motion parallax is an effective but incomplete decluttering solution for object recognition with visual prostheses.