Improvements to the Ladybug SDK v1.13
Subject
Technical Application Note : Improvements to the Ladybug SDK v1.13
Post Processing Features
Improved Default Settings
The default settings for the post processing pipeline have been improved.
Smear Detection
Greatly improved smear detection. Handles both unsaturated and saturated smear correction much better.
Independent Automatic Exposure
The ability to enable auto exposure for each sensor in the post processing pipeline was introduced. This replicates the same functionality as the traditional on-camera independent exposure mode.
Saturation
Saturation control for images is now supported. Allows users to boost or reduce the vibrancy of the images.
Levelling
Stretches the brightness levels in the image in order to improve brightness and contrast. Closely related to the white/grey point setting below.
White/Grey Point Setting
Provides the ability to override the default white and grey points as chosen by the post-processing pipeline.
Local Tone Mapping
New tone mapping feature that provides better image quality as compared to the existing OpenGL tone mapping feature.
Automatic White Balance
New feature to automatically detect the best white balance settings for the current image. White balance was previously restricted to 4 preset settings only.
Per-frame Post Processing Settings
Ability to define per-frame settings for post-processing. Available in both LadybugCapPro as well as directly through the Ladybug API.
Sample Images
Sample before and after images from Ladybug 1.7 SDK and 1.13 SDK.
1.7 SDK
1.13 SDK
GPU Acceleration
The following features have had GPU accelerated variants added:
- Local tone mapping
- Noise reduction
- False color removal
GPU acceleration is automatically utilized when a CUDA enabled graphics card is detected.
Results of GPU acceleration:
Operation | CPU Only Time |
GPU Accelerated Time | Speedup |
Process single image - All settings enabled |
21.9s | 1.4s | 15x |
Local tone mapping |
6s | 0.2s | 30x |
Noise reduction - CPU version is multithreaded |
1s | 0.03s | 33x |
False color removal |
7.9s | 0.1s | 79x |
The test was performed on a desktop PC with Core-i7-4770 CPU and Nvidia GTX 780 video card running Windows 8.1.
Other Improvements
Calibration Accuracy
The lens distortion model has been improved to provide more accurate rectified images. This allows more accurate mapping of image pixels to their physical location. On average, the Ladybug 5 now delivers an average angular error of 4mm at 10m distance.
Falloff Correction Quality
The falloff model has been improved to minimize differences across the stitched seams. This is especially visible on top seams that contain sky.
Linux Support
Introduced support for acquisition from camera to stream file, including GPS support. The API is also available for custom application development.