Flux Tensor Constrained Geodesic Active Contours with Sensor Fusion for Persistent Object Tracking

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Object tracking systems are crucial for surveillance and analysis.
• Existing methods struggle with complex scenes, varying illumination, and shadows.
• Multispectral data fusion offers potential for robust tracking.

Purpose of the Study:

• To develop an advanced object tracking system using novel motion detection and segmentation algorithms.
• To improve accuracy and robustness in challenging environmental conditions.
• To enable reliable tracking in multispectral video data.

Main Methods:

• Utilized a flux tensor algorithm for efficient motion detection in infrared video, independent of background modeling.
• Employed level set-based geodesic active contour evolution for object segmentation, fusing visible color and infrared edge information.
• Integrated a shape-based model for refining touching or overlapping objects.
• Extended multiple object tracking using correspondence graphs, Kalman filter-based cluster trajectory analysis, and watershed segmentation to handle occlusions and object groups.

Main Results:

• The flux tensor motion detector demonstrated higher accuracy than traditional methods in infrared video.
• The system proved insensitive to shadows and illumination variations across visible and infrared channels.
• The object tracking algorithm successfully handled challenging outdoor multispectral videos with occlusions and complex scenes.

Conclusions:

• The proposed flux tensor-based motion detection and multispectral fusion approach significantly enhances object tracking performance.
• The system offers a robust solution for real-world monitoring tasks under adverse conditions.
• This research contributes a more accurate and reliable object tracking system for diverse applications.