How Computers Track the Position and Angle of Multiple Objects

A method for using multiple cameras to track the exact location and orientation of many similar objects in a 3D space, even when they overlap.

What it covers

This patent describes a system that uses multiple cameras to observe a 3D space containing several similar objects. First, it identifies 'representative points' for each object in the camera images. It then calculates where those objects are in 3D space by comparing these points across different camera angles. Finally, it uses a pre-trained machine learning model to estimate the 'attitude'—or the specific angle and orientation—of each object by combining its calculated position with visual features like the shape of the object in the image.

What it doesn't cover

• —Does not cover systems that use only a single camera to estimate 3D position.
• —Does not cover object tracking methods that rely on depth sensors like LiDAR or Time-of-Flight cameras.
• —Does not cover tracking systems that require objects to have unique visual markers or QR codes.
• —Does not cover methods that do not use a regression model to estimate the final object orientation.

The clever bit

The system handles overlapping objects by assigning 'predetermined values' to features when images overlap, effectively telling the machine learning model to ignore the corrupted visual data and rely on the known spatial position instead.

Why it matters

In industrial robotics and automated manufacturing, tracking multiple identical parts (like screws or automotive components) moving on a conveyor belt is difficult. This technology helps robots 'see' exactly how an object is tilted or rotated, which is essential for picking it up correctly. It provides a mathematical way to handle the common problem where one object blocks another from the camera's view.

Real-world examples

• 1.Automated robotic assembly lines
• 2.Warehouse sorting systems for identical items
• 3.Multi-camera computer vision for manufacturing quality control