Moving Object Tracking Based on Sparse Optical Flow with Moving Window and Target Estimator

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Conventional moving object detection methods suffer from high computational costs, noise sensitivity, and target disappearance due to occlusions.
• Existing techniques struggle with real-world complexities like varied illumination and dynamic environments.

Purpose of the Study:

• To introduce a new moving object detection and tracking algorithm utilizing sparse optical flow.
• To address limitations of conventional methods by reducing computation time, enhancing noise reduction, and improving target estimation efficiency.

Main Methods:

• Developed a sparse optical flow-based algorithm incorporating a moving window detector for feature point selection based on location history.
• Implemented a memory-based estimator to retain corner feature locations, enabling tracking of obscured targets.
• Evaluated the algorithm on an embedded platform (Raspberry Pi 4) under diverse real-world conditions.

Main Results:

• The proposed algorithm significantly improves moving object detection performance.
• Achieved high frames per second (FPS) and enhanced accuracy compared to traditional optical flow, Haar-like, and HOG methods.
• Demonstrated effective tracking of targets even when obscured by obstacles.

Conclusions:

• The novel sparse optical flow algorithm offers a robust and efficient solution for moving object detection and tracking.
• The method shows superior performance in challenging environments with varying illumination, multiple objects, and occlusions.
• The developed approach is suitable for real-time applications on embedded systems.