Commercial Products

CoVA:
Comprehensive Vision Analysis

The Leading ML Networks.

3D Structure from Motion.

Custom Data Flows. 

 

Seamlessly integrated in a single workspace:  CoVA

Novel and powerful image processing algorithms are delivered almost daily by researchers around the world. To capitalize on these advancements and use them together seamlessly, Knight developed the Comprehensive Vision Analysis tool: CoVA. Our CoVA software significantly reduces the time and effort to apply advanced processing to images and video and is a core component of several of our commercial products. CoVA provides each ML network with the necessary runtime environment, enabling complex processing pipelines to be built on a single platform. These pipelines can then be run locally for development, or in the cloud for massive parallelization.  CoVA brings all your models to a single workspace:

Polar Ice Segmentation for Navigation Planning

The power of CoVA is on display in this rapid application of modern segmentation and tracking tools to a novel problem: Ice Tracking for Navigation.  Using a combination of off-the-shelf algorithms and custome processing, Knight can deliver automated segmentation and tracking form a variety of ice concentration data sources.

Animal Pose Measurement

Automated pose estimation applied to animals is the key to improving the accuracy and throughput of animal observations, be they for research, livestock, or family pets.  Knight’s custom processing succeeds where off the shelf networks fail.

Markerless Pose Capture

Knight develops several commercial products using several interlocking technology areas. Our research into Markerless Pose Capture has resulted in a deployable, modular MPC system using commercial cameras and custom processing. With this capability, we can produce dimensionally accurate 4D human poses. Extensions to animal models require only the substitution of animal pose estimation routines. Knight uses open-source tools, modern machine language networks, commercial camera systems, and proprietary image processing algorithms to provide in-situ markerless pose capture. Our MPC setup can reconstruct dimensionally accurate human poses at 120 frames per second.

3D Models from Full Motion Video

Combined with the proliferation of imaging sensors across the world, structure from motion approaches provide the ability to generate three dimensional models for vehicles, buildings, and landscapes of interest.  Knight Scientific Systems uses modern processing approaches to deliver 3D models from photographs and videos.

  • Visual Odometry and Photogrammetry
  • SVO
  • REMODE
  • Gaussian Splatting
  • NERF
  • SLAM

Multi-Spectral Object Detection

Traditional Object Detection (OD) algorithms are trained only on visible imagery, meaning the ability to detect vehicles, people, signs, etc., is only mature for visible imaging systems.  Because Mid-Wave InfraRed (MWIR) and Long-Wave InfraRed (LWIR) sensors play an important role in detection and surveillance missions, Knight is developing multi-spectral OD tools to bring the power of machine learning to all imaging wavebands.

Machine Learning Enhanced Detection, Tracking Identification and Characterization

Classical approaches to processing intelligence, surveillance and reconnaissance (ISR) sensor data are significantly improved when enhanced with modern Machine-Learning technologies like feature detection and classification. Our team is at the forefront of research and development of AI/ML applications to military missions. Our AI/ML activities include the following processes and tools.

  • Data Collection and Feature Extraction
  • ImageDet
  • YOLO
  • Tensor Flow
  • PyTorch