Image creating and sharing continues to rise rapidly in the whole world and demands new approaches in image processing. Huge amount of different types of devices (cams, lenses) leads to situation when there are no single solutions even for the same tasks (distortion, denoising and etc.). Aligned Research team uses advanced technology and algorithms to process image data fast and within appropriate quality metrics. Our primary interest is in reinforcement of math methods of image analysis for solutions in augmented reality, computer vision, pattern recognition, image preprocessing and correction, and automatic 3D reconstruction from multiple images.
Advantages of Aligned Research Group technology are as follows:
- GPU optimization
- Competitive performance
- Affordable licensing
There are two common ways to perform 3D reconstruction – a parallax-based method to infer geometric structure of a scene captured with a collection of images and object segmentation using thin 2D slices. The latter approach is widely used in medical diagnostics in MRI and CT scanning.
During 3D reconstruction it is possible to perform object segmentation. This is frequently used in organ segmentation using DICOM data from CT. In the augmented reality surgery assistance project we work in collaboration with the medical school, the application we are building will be able to produce precise multicomponent anatomical model based on noisy and low-quality DICOM data. This model will then be used for pre-operative planning and to create an effect of augmented reality for the videostream of the surgical field.
3D reconstruction technology is available to be included as a module into your project, or to be customized to your requirements.
Video Stream Stitching
Still image stitching is widely used and an integral part of many editing software. Inexpensive but high quality video cameras becoming widespread and frequently used to capture streams from adjacent areas. With video stream stitching, it’s possible to combine output of multiple cameras into one aggregated, high resolution stream. Unlike still image panoramic stitching, which can be done in batch mode, video stitching must frequently be done in real time. This demands robust methods of distortion suppressing and control points discovery. We continuously refine our calibration method and work on reducing computation complexity of our stitching code. GPU optimization is performed for the most critical code paths in our images processing algorithms.
Image Pre-processing and Correction
Image correction is one of those invisible tasks that is required for higher-level tasks but frequently ignored or left to automatic “auto-levels” approach. When done right, image correction can dramatically improve results of 3D reconstruction, stitching, OCR, computer vision and pattern recognition.
Image correction is a broad term, that includes a variety of distinct steps, which include:
- exposure correction
- color and white balance correction
- lens distortion correction
- perspective control and other geometry distortion correction
We use a combination of in-house algorithms and improved open source libraries. Aligned Research Group scientists and engineers continuously work on improving the quality of our algorithms, as well on porting and optimizing them. For real-time applications, we ported the most time-consuming algorithms to utilize NVIDIA GPU and Intel Phi.