On its face, change detection seems to be the easiest thing in the world. Take two images of a scene – a before-image and an after-image – align them and subtract them. Where the difference image is nonzero, you have change. The problem, of course, is that this gives you a change at every pixel in the image, and the vast majority of changes found in this way are not significant. Detected changes can be due to differences in viewpoint, aspect, and scale; illumination, which has both spectral (e.g., white balance) and spatial (e.g., shadows) attributes, focus, resolution, sensor calibration, automatic gain, and state of health; atmospheric absorption, emission, and scattering, etc. In addition, there are real changes in the scene, which may or may not be meaningful, depending on the application: moving of cars and trucks on the roadway, growth and senescence of vegetation, clouds in the air, etc. Change detection is a particularly appealing approach for many intelligence problems, because they are, by design, difficult to see just by themselves, but often other indicators provide information necessary for detection.
Highly accurate image registration is the critical piece in any change detection process. When images are properly registered, valid changes are emphasized and false alarms reduced. Accurate registration also facilitates automated processing and ensures more timely, robust, and accurate analyses.
Observera’s Sensor Model-based Automated Registration Tool (SMART) enables automated pixel level registration and the generation of cueing layers. Our approach to automation incorporates sound photogrammetric principles, automated tie point identification, and robust optimization procedures resulting in pixel level registration between images. This automated change detection workflow can alleviate tedious computational tasks that do not require a lot of human judgment. It can also shorten the timelines required for training operators.
Observera’s change detection technology is unique in three primary ways: (1) it uses our proprietary registration library to automatically generate registered image pairs using sound photogrammetric principles; (2) once registered a parallax correction is performed to help eliminate errors due to elevation and viewing geometries; and (3) it incorporates advanced resampling and filtering techniques that make it reasonably robust to differing time-between-collects, imaging geometries, illumination directions, and environmental conditions.
Contact us for more information about EOCD.
* Sensor Model-based Automated Registration Tool
Key Features
- Works on imagery of differing orientation
- Handles different illumination & environment
- Works with & without elevation info
Solution
- Image-to-image registration with SMART*
- Perform fine-grained parallax correction
Applications
- Large Area Search
- Facility Monitoring
- Construction Chronologies
- Vehicular Activity
- Activity Detection & Monitoring
- Construction Progression Analysis
- Object Counting
- Increased Imagery Review Speed
- Detection of Detailed Changes
Contact us: