UnCanny: Exploiting Reversed Edge Detection as a Basis for Object Tracking in Video.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Existing object detection methods struggle with small objects (<20 pixels) in complex backgrounds.
• High computational cost is a limitation for real-time applications.

Purpose of the Study:

• To develop a computationally inexpensive framework for detecting small objects in static, complex backgrounds.
• To enable robust object tracking from sequential video frames.

Main Methods:

• A novel framework reversing classic edge detection steps.
• Utilizes Canny filter for edge detection, combined with image subtraction, thresholding, Sobel edge detection, Gaussian blurring, and Zhang-Suen edge thinning.
• Processes sequential video frames to identify inter-frame object movement.

Main Results:

• The framework successfully identifies distinct contours of moving objects.
• Achieves object detection with minimal false positives.
• Demonstrates applicability to object tracking algorithms.

Conclusions:

• The proposed framework offers a robust, low-overhead solution for small object detection and tracking.
• It can be integrated with various edge detection methods for enhanced performance.
• Potential for real-time applications with complex visual data.