3D Machine Vision Uses Single Pixel for High Speed and Cost Efficiency

Self-driving cars rely on advanced sensors and significant computing power. However, a research team led by Tsinghua University in China has developed a tracking system that reduces computing requirements to just a single pixel.

Initially, the computer vision systems required for autonomous vehicles and intelligent security seemed straightforward: connect a camera to a processor, add a few algorithms, and you’re set. Humans manage complex image processing effortlessly with their biological systems, so how challenging could it be?

Challenges in Emulating Human Vision and Reflexive Response

It turns out this is extremely difficult. While the human eye is optically simpler (though still complex), the underlying processing is incredibly intricate. Moreover, humans use this information and respond reflexively, making it nearly impossible to emulate.

The Tsinghua research team has made a breakthrough by creating a 3D tracking method for fast-moving objects at exceptional speeds, all while lowering computing demands. The innovation lies in achieving this not with a full image, but by utilizing just a single pixel.

Efficient Object Tracking Without Image Reconstruction

“Our method doesn’t require reconstructing the object’s image to determine its position, greatly reducing data storage and computational costs,” explained research leader Zihan Geng.

“Specifically, calculating a 3D coordinate needs just six bytes of storage and 2.4 microseconds of computation time. By lowering computational expenses and enhancing efficiency, it could decrease the cost of equipment for high-speed tracking, making the technology more affordable and opening up new applications.”

The team claims their technique can track objects 200 times faster than conventional video-based methods using just one detector, with no prior information and minimal computing power.

It achieves this by projecting geometric light patterns onto the object and measuring the intensity at a single pixel. Through complex mathematical calculations, the object’s position and trajectory are determined.

Using a Laser and Single-Pixel Data for Object Tracking

Initially, the team used simulations, but they soon progressed to using a laser and a digital micro-mirror device (DMD) to illuminate a metal sphere moving along a curved spiral wire. The system performed calculations based on the data from just one pixel.

Currently, this method tracks a single object, with the next goal being to track multiple objects simultaneously.

“This technology could enhance perception in self-driving cars, improve security surveillance systems, and provide more efficient monitoring and quality control in industrial inspections,” Geng stated. “Moreover, this high-speed localization technique can be applied in scientific research, such as studying the flight paths of insects.”

Read the original article on: New Atlas

Read more: Particle Accelerator Tech from CERN now Treats Brain Tumors