How Cameras Automatically Re-Adjust Themselves Using Fixed Objects

This patent describes a method for surveillance cameras to automatically correct their aim and focus by comparing what they currently see to old calibration images, specifically looking for fixed objects in the scene.

What it covers

The patent outlines an automated system for recalibrating cameras that monitor a real-world area. It works by taking new images and comparing them to a set of older, known-good calibration images. A special computer program, called a trained neural network classifier, finds specific points on 'relatively immobile structures' (like walls or furniture) in both the new and old images. If the way these points have shifted, called 'transformation information,' reaches a certain level, the camera's settings are automatically updated to correct its view. For example, if a security camera slightly shifts due to vibration, this system would detect the shift by comparing a new image of a doorway to the original calibrated image of that same doorway, then automatically adjust the camera's internal settings to restore its proper alignment.

What it doesn't cover

• —Does not cover manual recalibration where a human physically adjusts the camera or its settings.
• —Does not cover recalibration methods that rely on moving objects or people in the monitored area.
• —Does not cover systems that recalibrate without using a trained neural network classifier to identify features.
• —Does not cover recalibration that does not compare current images to a previously stored set of calibration images.
• —Does not cover recalibration based on changes in lighting or color, focusing instead on structural features.

The clever bit

The clever part is using a trained neural network to specifically identify and track features on 'relatively immobile structures' within an area. This allows the system to accurately detect camera drift and automatically correct it, without needing special markers or human input.

Why it matters

Automated recalibration is important for any system relying on consistent camera views, such as security, robotics, or industrial monitoring. It reduces the need for human intervention, saving time and labor costs. This technology helps ensure that cameras remain accurate over long periods, even if they experience minor shifts or disturbances.

Real-world examples

• 1.Smart home security cameras
• 2.Industrial monitoring systems
• 3.Robotics vision systems
• 4.Traffic monitoring cameras
• 5.Augmented reality applications that map real spaces