Depth from a Motion Algorithm and a Hardware Architecture for Smart Cameras.

Area of Science:

• Computer Vision
• Robotics
• Embedded Systems Engineering

Background:

• Depth map estimation is crucial for autonomous systems (navigation, robot vision, flying).
• Traditional depth from motion algorithms suffer from low processing speeds and high hardware demands, limiting embedded applications.
• Existing optical flow methods often lack the efficiency and accuracy required for real-time performance.

Purpose of the Study:

• To propose a novel hardware architecture for efficient depth from motion estimation.
• To develop an improved optical flow algorithm suitable for embedded systems.
• To achieve high-accuracy, dense depth maps with significantly enhanced processing speeds.

Main Methods:

• A new optical flow algorithm extending stereo matching using a pixel-parallel/window-parallel approach with Sum of Absolute Difference (SAD).
• Incorporation of the curl of the intensity gradient as a preprocessing step to enhance SAD performance.
• Development of a hardware architecture integrating flow/depth transformation and the proposed optical flow algorithm.

Main Results:

• The proposed optical flow algorithm achieved 90% accuracy, outperforming previous Field Programmable Gate Array (FPGA)-based methods.
• The system generates dense depth maps with motion and depth information for all image pixels.
• Processing speed is up to 128 times faster than previous approaches, enabling real-time embedded applications.

Conclusions:

• The developed hardware architecture and optical flow algorithm offer a significant advancement in depth from motion estimation.
• The solution addresses the limitations of traditional methods, providing high accuracy and speed for embedded systems.
• This work facilitates improved performance in applications requiring real-time depth perception.