Toward wide-angle microvision sensors

Area of Science:

• Computer Vision
• Microscale Devices
• Optical Engineering

Background:

• Microscale devices face severe power and mass constraints, challenging common computations like matrix manipulations and convolutions.
• Existing computer vision solutions are often too power-intensive or bulky for microscale applications.

Purpose of the Study:

• To propose and analyze a class of miniature vision sensors designed to overcome power and mass limitations for microscale computer vision.
• To enable efficient computer vision tasks on resource-constrained platforms.

Main Methods:

• Development of miniature vision sensors utilizing template-based optical convolution to reduce power requirements.
• Implementation of a refractive optical design to achieve a wide field-of-view within a compact form factor.
• Analysis of design tradeoffs including field-of-view, volume, and mass, supported by analytic tools.

Main Results:

• Demonstration of milliscale prototypes capable of computer vision tasks such as edge localization, target tracking, and face detection.
• Successful miniaturization of optical designs using photolithographic fabrication.
• Onboard fiducial detection demonstrated on a small autonomous air vehicle.

Conclusions:

• The proposed miniature vision sensors effectively address power and mass constraints in microscale computer vision.
• The developed sensors and design methodologies enable practical computer vision applications on small, autonomous platforms.
• Photolithographic fabrication offers a viable path for further miniaturization and integration of these vision systems.