EDGE20: A Cross Spectral Evaluation Dataset for Multiple Surveillance Problems.

Area of Science:

• Computer Vision
• Machine Learning
• Surveillance Technology

Background:

• Existing surveillance datasets often focus on single tasks and visible spectrum (VIS) cameras.
• Previous cross-spectral datasets were acquired under constrained conditions, limiting real-world applicability.
• Unconstrained outdoor environments present significant challenges for current detection and recognition methods.

Purpose of the Study:

• Introduce the EDGE19 dataset for robust pedestrian detection, face detection, and face recognition.
• Enable research on cross-spectral analysis using visible (VIS) and near-infrared (NIR) spectra.
• Provide a benchmark for evaluating algorithms under unconstrained, real-world conditions.

Main Methods:

• Collected images using trail cameras in outdoor environments during day and night.
• Acquired data under unconstrained conditions, including variations in pose, illumination, and motion.
• Annotated the dataset with bounding boxes for pedestrians and faces, unique subject identifiers, and facial pose labels.

Main Results:

• Evaluated the performance of state-of-the-art methods on the EDGE19 dataset.
• Baseline results indicate significant challenges for current methods in cross-spectral tasks.
• Identified key difficulties including low resolution, pose variation, illumination changes, occlusions, and motion blur.

Conclusions:

• The EDGE19 dataset provides a valuable resource for advancing cross-spectral surveillance research.
• Current methods struggle with unconstrained, real-world conditions, highlighting the need for improved algorithms.
• Future research should focus on developing more robust methods for diverse environmental challenges in surveillance.