a computer vision R&D consultancy, marrying state of the art algorithms with high-performance computing, paying special attention to exceptionally high standard software quality.
Around every real-life computer vision problem there is a large amount of domain knowledge. Integrating this into the state-of-the-art algorithms makes a real difference in the final success of the solution. Our experts have in-depth knowledge of modern algorithms and know how to extend to use all of this special knowledge to arrive at industry-leading solutions. Instead of plug-and-play with off-the-shelf algorithms we create, redesign, and combine state of the art to get you the best possible performance. Our computer vision core libraries, experience, and know-how are a guarantee for success.
Software performance is crucial when it comes to demanding business requirements or engineering challenges such as industrial computer vision. More so, as it directly results in a better product from the customer's perspective. We never heard the term too fast when it comes to these solutions. We have proven to be capable of improving performance on all levels, whether the critical loop is measured in microseconds or days of computing. We have experts skilled in algorithmic speed improvements as much as we have them in low-level hardware-dependent optimizations.
Poorly written software usually belongs to the programmer a lot more than the company who paid for it. We believe software design and quality is crucial for a successful business with long-term plans. When working with old, poorly written code, the programmer sacrifices significant mental energy to understand, and cope with the decisions made by the previous developer. This is the reason most of the expensive software rewrites happen: the new programmer finds it less painful. Our code comes with the highest quality standards, and careful software structures designed for intuitive understanding and reusability. Our software lives long.
Spectral MD Inc. (SMD) owns a proprietary multispectral imaging technology that is combined with AI / Deep Learning algorithms for faster and more accurate treatment decisions in wound care. Their imaging technology is based on their custom multi-camera rig. For their AI-driven analysis, 3D reconstruction and the ability to align simultaneous captures to each other is crucial.
Working with SMD, we significantly improved the 3D reconstruction pipeline achieving higher quality reconstructions while keeping the computation time intact.
We are experts in 3D reconstruction that we demonstrated in many of our projects. Combining our knowledge of computer vision and high-performance optimization techniques, we developed a more than 10x faster multi-view stereo algorithm for our partners to enable less than one day scan processing turnovers from the original 3 days.
Our experience also includes efficient CPU/GPU algorithms for point cloud operations and filtering. These point clouds are often large and complex, requiring significant computational resources to process and analyze. There are many point cloud software libraries helping common operations, however, most of them uses significant performance compromises to be general purpose. In the industry it is common that your point cloud does not fit in main memory along with additional data such as RGB images, segmentation masks, normal and depth maps. With domain-knowledge driven algorithms and software techniques we are able to filter noisy and incomplete point cloud data while remaining memory and computation efficient.
Free Space Optical Communications is a field that is continuously gaining attention for secure, high-bandwidth data transmission ground-to-link satellite, and ground-to-ground communication where physical connections are impractical. One of the main challenges related to this field is the turbulence characteristics of the atmosphere that deform the wavefront of the light and deteriorate communication. This deformation varies at a rate in the order of 100-200Hz.
Aconic is responsible for designing, developing, and optimizing a real-time control system to compensate for the deformations using adaptive lenses. In the project, the maximally allowed computational latency is 50us after the image acquisition, making the data structure and low-level optimizations especially difficult.
All of the 3D reconstruction algorithms will fail if the camera's internal parameters and their position are not calibrated precisely enough. At Aconic, we can provide subpixel-level calibrations even in very extreme conditions. Over the years we have experimented and built many calibration algorithms and optimization techniques with which we were able to accurately calibrate even using only a few images.
The aconic.ai team consists of several experts perfectly complementing each other with overlapping experience and interests. Each team member is multidisciplinary in their craft, and highly passionate about their work.
