Real-time object tracking based on scale-invariant features employing bio-inspired hardware.

Area of Science:

• Computer Vision
• Robotics
• Embedded Systems

Background:

• Scale-Invariant Feature Transform (SIFT) is crucial for object recognition and image matching.
• Real-time implementation of SIFT is computationally demanding for traditional systems.
• Efficient hardware acceleration is needed for advanced vision tasks.

Purpose of the Study:

• To develop a novel vision sensor system capable of performing SIFT in real time.
• To optimize SIFT algorithm execution through parallel processing techniques.
• To demonstrate the system's effectiveness in feature point tracking applications.

Main Methods:

• The system integrates an active pixel sensor, a metal-oxide semiconductor (MOS)-based resistive network, and a field-programmable gate array (FPGA).
• Whole-image parallel filtering is achieved using the MOS resistive network with configurable filter sizes.
• Frequency-band parallel processing is implemented via pipelining on the FPGA.

Main Results:

• The developed vision sensor system successfully performs scale-invariant feature transform (SIFT) in real time.
• The parallel filtering and processing architecture significantly enhances SIFT computation speed.
• The system demonstrated effective tracking of feature points on an object within a video sequence.

Conclusions:

• The proposed vision sensor system offers an efficient hardware-based solution for real-time SIFT.
• This approach enables high-speed feature detection and tracking for dynamic vision applications.
• The integration of specialized hardware components like MOS networks and FPGAs is key to achieving real-time performance.