Adaptive tunneling photodiodes enable visual recognition in high-contrast scenes

Area of Science:

• Computer Vision
• Materials Science
• Electrical Engineering

Background:

• Accurate visual sensing is crucial for intelligent systems like autonomous vehicles and robotics.
• High dynamic range (HDR) imaging is essential for capturing scenes with extreme illumination variations.
• Current HDR solutions face limitations such as latency, motion artifacts, and high power consumption.

Purpose of the Study:

• To present a hardware-centric approach for achieving HDR functionality directly within camera image sensors.
• To overcome the limitations of existing computational and hardware-intensive HDR techniques.
• To enable robust visual sensing in challenging high-contrast lighting environments.

Main Methods:

• Developed a camera image sensor with integrated HDR functionality at the photodetector level.
• Utilized an engineered tunneling mechanism for bias-controllable, continuously tunable dynamic range.
• Tested the sensor-level HDR capability in real-world autonomous driving scenarios.

Main Results:

• Achieved a continuously tunable dynamic range of up to 150 dB.
• Enabled robust object classification with confidence scores exceeding 91% in high-contrast autonomous driving scenes.
• Demonstrated substantial reduction in computational load by shifting HDR processing to the sensor front-end.

Conclusions:

• The sensor-level HDR approach significantly enhances system-level efficiency, speed, energy consumption, and recognition reliability.
• This hardware-centric solution paves the way for next-generation, compact, on-chip visual preprocessing hardware.
• The developed prototype offers a promising solution for intelligent vision systems requiring advanced visual sensing capabilities.