Skip to main content
0

Revisualize Radar with Radar Vision

Radar technology has existed for a long time. Over the years, innovations have taken place that have evolved Radar into a very capable sensor albeit with significant limitations. One such limitation is that its resolution is quite low, especially compared to optical and light-based sensors. For this reason, we feel that Radar has been overlooked as a sensor in the modern marine industry, especially recreational marine – Radar is a standard for collision avoidance, but its output is clunky and requires experience and training to interpret.

Since 2021, Tocaro Blue has been testing and developing the ProteusCore™. The Core is a machine learning (a form of Artificial Intelligence) pipeline that ingests Radar scan lines, and outputs a machine readable list of classified and tracked objects. The relative simplicity of this system marks a significant leap forward in Radar technology, and is a new basis for human interface focused displays. The ProteusCore™ can determine whether an object is a small boat, medium boat, or large boat, or a variety of buoys and markers, with confidence and complete automation.

With the ProteusCore™ your Radar, MFD, or autonomous control system has the power of Radar Vision.

Below are the primary components that make up the ProteusCore™ software suite. The ProteusCore™ software is available for purchase as a software license from Tocaro Blue to empower your Radar, MFD, Navigation Software, or autonomous control system.

Radar Auto-focus

ProteusCore™ Auto-focus eliminates the clutter that is found in traditional RADAR data using a machine learning (ML) filtering process that treats RADAR like a video stream and identifies real targets in the image. Typically we find that 85 to 90% of the contents of each RADAR scan are clutter that can be ignored – Proteus focuses the visualization and tracking on the remaining seaborn objects that really matter. The ML model sends feedback controls to RADAR to tune it in real-time to achieve optimal target detection.

Machine Learning Target Classification

After the Auto-focus has removed the clutter and detected objects, ProteusCore™ classifies the seaborn objects it detects to determine how they will behave. For instance, jetskis are quick and nimble – they can accelerate quickly, becoming a threat at any time. At the other end of the spectrum, barges have significant momentum and follow very predictable paths. Buoys can be displaced when tide currents are flowing. Class information is used to track targets more accurately and provide better predictive power.

Predictive Power

Our high-confidence classification of targets and sensor fusion of your onboard sensors allows us to accurately simulate our own path and other vessel paths up to 30 seconds into the future. This simulation informs our at-a-glance Field of Awareness display, which keeps you informed of important closest-points-of-approach (CPA) with other vessels.

Sensor Bias Correction

Did you know typical auto-pilot grade compasses are only accurate to within a few degrees? Even this accuracy can be thrown off by 5 degrees or more when near ships or bridges (that’s an error of 140 yards for a target at one mile range)! As ProteusCore™ combines data from your onboard navigation sensors with focused RADAR imaging, it is able to detect discrepancies between the sensors, model results, and cartography then amend these errors before providing results to the captain.

Training Data

How do we train the machine learning model? Between our test boat Radar Love and our fleet of Early Access Partners, Tocaro Blue is logging hundreds of hours of data each month in locations all along the Eastern Coast of the United States. We have collected data in all sorts of weather, in narrow canals and busy harbors – every new scenario is used to improve the reliability and robustness of the ML models. The test fleet operates on a wide range of RADAR (both magnetron and solid-state) and OEM equipment to validate performance on many recreational boats.