Filed Jun 7, 2021
What this patent covers
A method for understanding objects and predicting their movement using camera imagery—focused on moving objects (vehicles and other transportation) and their environment (roads, buildings, parking areas).
The problem
Most camera-based systems rely only on raw neural-network outputs. That typically produces incomplete and imprecise data, limiting reliability for real-world prediction.
Our approach
We convert image-derived detections into a vector-based object model, store it in a database, and use it to build a controlled 3D digital replica of the observed space.
A physics-based simulation layer is then used for realism and consistency checks—improving accuracy and stability in prediction.
Why it matters
• More reliable 3D understanding from camera imagery
• Stronger tracking and prediction in real-world scenarios
• Better handling of constraints such as position, motion, and plausible changes in the scene