Reflection Removal Under Fast Forward Camera Motion.

Area of Science:

• Computer Vision
• Image Processing
• Video Enhancement

Background:

• In-vehicle camera footage quality is often compromised by windshield reflections, dirt, and dust.
• These artifacts degrade the clarity of dashcam recordings, impacting their utility.

Purpose of the Study:

• To develop a novel algorithm for simultaneous removal of reflections and dirt artifacts from dashcam videos.
• To address image degradation issues caused by internal reflections and external contaminants on the windshield.

Main Methods:

• Exploiting spatiotemporal coherence of stationary reflections and dirt against a fast-moving background.
• Proposing an average image prior to enforce spatiotemporal coherence.
• Utilizing a two-layer separation model (stationary and background layers) with distinct gradient sparsity distributions.
• Implementing motion compensation in postprocessing to mitigate vehicle vibration-induced layer jitter.

Main Results:

• The proposed algorithm successfully separates stationary reflection and dirt layers from the background.
• Demonstrated effectiveness on both real and synthetic dashcam video datasets.
• Achieved simultaneous removal of reflections and dirt artifacts under fast camera motion.

Conclusions:

• The novel algorithm effectively enhances in-vehicle black box camera video quality.
• The method provides a robust solution for de-reflection and artifact removal in challenging driving conditions.
• The approach offers improved clarity for critical video evidence captured by dashcams.