anomaly-detection-vision

Detect scene changes by comparing current camera frames against a stored baseline reference image. This visual anomaly detection identifies moved objects, opened doors, or missing items to support reliable surveillance and patrol monitoring.

Comparing camera frames to a reference image is best handled through configurable methods like pixel-difference, SSIM, feature-based matching, and embedding-based similarity. These techniques generate a robust anomaly score for security monitoring.

Yes, visual anomaly detection is designed for patrol robots and warehouse security. It reliably detects unusual activities like new items or moved objects by establishing baselines and monitoring camera feeds for environmental changes.

Manage reference images by loading a stored baseline for a waypoint and starting vision-based anomaly monitoring. The system supports reference management to allow periodic updates and maintain reliable detection over time.

Temporal filtering provides robust anomaly detection by smoothing out transient visual noise. Combined with behavior tree integration, it ensures reliable automated surveillance and reduces false positives when detecting scene changes.

Pixel-difference comparison alone can be sensitive to lighting changes, which is why SSIM, feature-based matching, and embedding-based similarity are also available. These combined methods provide a more robust and configurable anomaly score.