Humans are such visual beings that we tend to forget that computers are blind. We unconsciously assume all intelligent systems can perceive the world like we do. The holy grail of sensor processing and networking is to provide intelligent systems with the functional equivalent of human vision, but that goal is still a long way off. Charles River Analytics’ Sensor Processing and Networking Division focuses on practical sensor processing and networking problems that are well defined and where basic research in academia has already made significant progress. We bring this research from the lab to our customer’s systems, in the form of practical algorithms, software, and integrated systems.

Our approach to sensor processing and networking is focused in four primary areas:

• Automatic/aided target recognition (ATR). Soldiers and image analysts routinely use advanced sensors of all types (infrared imaging, laser-radar, vibration, etc.) to detect and recognize enemy targets. But in the heat of battle, intelligent systems are desperately needed to help people process the deluge of images. We develop feature- and model-based ATR software, as well as a variety of tools for other ATR developers. For example, our START software for “truthing” large video data sets enables ATR developers to compare—rapidly and cost effectively—their ATR system’s performance with that of a hypothetically perfect ATR system.  
• Security and surveillance. Cameras are everywhere, but who is looking at all the images? Sensor processing and networking systems enable computers to do some of the looking. The human visual system is enormously capable, but people tire and get distracted; a sensor processing and networking system provides consistent performance 24/7. We develop systems for pedestrian/vehicle detection and tracking, event recognition (“person exiting car,” etc.), license plate detection, and 3D reconstruction of an object from its image.
• Vision-based navigation. Mobile robots of all kinds (air, ground, water surface, and underwater) rely on a human operator for navigation, generally through direct teleoperation. We develop a variety of vision system components for increased robot autonomy, such as navigating by matching flyover imagery with onboard maps, urban route-finding via sign recognition, and path planning via obstacle detection. We’re even working with the Navy to develop a UAV that can taxi on an aircraft carrier by recognizing the arm gestures of a flight-deck director.
• Image and video enhancement. Most images captured with electro-optic sensors need enhancement of some sort before being viewed by a person or processed by an automated system. We develop software that recognizes what kind of artifacts/noise are degrading the imagery, and then applies the needed image enhancement techniques that best recover the image. We’re doing this for real-time applications, such as fixed-pattern noise and dynamically varying noise in video, and for more computationally complex applications, such as document imaging.

The Sensor Processing and Networking group at Charles River Analytics is a talented team of researchers and software developers with great experience in building custom sensor processing and networking systems. We have developed a variety of solutions in the application domains listed above (and a few related domains), ranging from basic research in algorithm design to highly optimized end-to-end systems. Along the way, we have built up an extensive in-house library of sensor processing and networking software components (VisionKit) which allows us to rapidly prototype and evaluate new algorithms. Unlike dedicated algorithm prototyping environments (such as MATLAB™), our library is written in a high-performance compiled language (C++) and strictly engineered for real-time performance.

Example Projects: