Consecutive Image Acquisition without Anomalies.

Area of Science:

• Computer Vision
• Robotics
• Machine Learning

Background:

• Images acquired by cameras on moving platforms (rovers, drones) can have anomalies.
• Overlap anomalies occur when images have excessive overlap, while gap anomalies occur when images have insufficient overlap.
• These anomalies hinder the effective use of image data in subsequent analysis, particularly for machine learning.

Purpose of the Study:

• To introduce a novel algorithm, OVERGAP, for detecting and correcting image acquisition anomalies.
• To ensure the generation of consecutive, anomaly-free image sequences from moving camera platforms.
• To facilitate the use of corrected image data in machine learning processes for feature prediction.

Main Methods:

• The OVERGAP algorithm utilizes Dynamic Time Warping (DTW) distance and Wasserstein distance for anomaly detection and correction.
• It processes image sequences acquired from on-board cameras on moving vectors.
• The algorithm corrects both overlap and gap anomalies to produce images of a desired, consistent size.

Main Results:

• OVERGAP successfully identifies and rectifies overlap and gap anomalies in image sequences.
• The algorithm generates a stream of consecutive, anomaly-free images.
• The corrected images are suitable for direct integration into machine learning pipelines, especially Deep Learning models.

Conclusions:

• The OVERGAP algorithm provides an effective solution for image acquisition anomalies in dynamic scenarios.
• By producing anomaly-free image data, OVERGAP enhances the reliability and efficiency of machine learning models.
• This work contributes to improving data quality for visual information processing in robotics and autonomous systems.