Arcturus Achieves Global Recognition with Best In-Class Multiple Object Tracking Performance

Arcturus Networks Inc., a leading provider of edge AI analytics and enablement announced the results of its public submission to the global Multiple Object Tracking (MOT) benchmark competition where Arcturus Brinq Traq (BQTQ) achieved a MOTA benchmark score of 77.2 for MOT16 and 77.7 for MOT17 challenges. The results represent a 0.7 point gap from the top MOT16 position at 77.9 and a 2.6 point gap from the top MOT17 position at 80.3. This ranks Brinq tracking solutions in the top-10 globally when compared with all submissions (including leading research) and as the number one, commercially available edge solution.

Latest Aithority Insights: Detecting, Addressing and Debunking the Hidden AI Biases

Multiple object tracking remains an area of highly active research among academic institutes and commercial organizations with advancements in ML enabling powerful new methods. Multiple object tracking requires a unique ID to be generated for each object in the field of view. This anonymized ID is maintained frame-over-frame as the object moves, ultimately describing how each unique object interacts with the space around it. Accurate multiple object tracking is achieved by maintaining each unique track ID, a task referred to as reidentification. The advancement Arcturus ML data scientists achieved is increasing reidentification accuracy, while improving overall efficiency. The result is that Brinq tracking software is more tolerant to occlusions and performant on low-power edge devices.

“Tracking is a primitive for many applications as it provides critical insights beyond object detection. One novel application of tracking we’ve seen is in food handling, to ensure an employee uses the hand wash station prior to entering the food preparation area” said David Steele, director of innovation at Arcturus. “Other applications include healthcare, smart retail stores and public safety”.

AI and ML News: AI: Continuing the Chase for Brain-Level Efficiency

Brinq Tracking Solutions
Brinq provides a suite of tracking options, targeting three performance tiers.

• A low-complexity motion-prediction only method to address low-power or high-frame rate applications
• A medium-complexity motion-prediction solution that maintains tracklet data to achieve high-accuracy tracking even in situations with high object density
• A method that enhances either above approach with a visual appearance embedding, making it possible to re-associate an object with a track, when the motion track is lost

In addition, two enhancements are currently under development – Multi-Target, Multi-Camera (MTMC) tracking and long-term reidentification, each of which enhance tracking capabilities in either space or time domains.

AI ML in Marketing: AI and Big Data Analysis Used to Find Brands’ Emotional Connection

[To share your insights with us, please write to sghosh@martechseries.com]