Deep Learning-Based Target Tracking and Classification for Low Quality Videos Using Coded Aperture Cameras.

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Compressive sensing (CS) is increasingly applied in various fields.
• Pixel-wise Code Exposure (PCE) cameras offer low power consumption and individual pixel control.
• Current PCE camera applications require time-consuming, lossy frame reconstruction for analysis.

Purpose of the Study:

• To develop a deep learning framework for direct target tracking and classification in the compressive measurement domain.
• To eliminate the need for traditional frame reconstruction in PCE camera data processing.
• To improve the efficiency and practicality of PCE camera applications.

Main Methods:

• Utilizing the You Only Look Once (YOLO) algorithm for target detection and tracking within compressive measurements.
• Employing a Residual Network (ResNet) for direct classification of targets in the compressive domain.
• Validating the approach using low-quality optical and mid-wave infrared (MWIR) videos from the SENSIAC database.

Main Results:

• Demonstrated the efficacy of the proposed deep learning approach in direct target tracking and classification.
• Achieved successful detection and classification without the intermediate step of frame reconstruction.
• Validated performance on diverse video data, including challenging low-quality optical and MWIR imagery.

Conclusions:

• The proposed deep learning method enables efficient, direct target tracking and classification for PCE cameras.
• Eliminating frame reconstruction significantly streamlines PCE camera data analysis.
• The approach shows promise for practical applications utilizing PCE camera technology.