Automatic Dynamic Range Adjustment for Pedestrian Detection in Thermal (Infrared) Surveillance Videos

Area of Science:

• Computer Vision
• Image Processing
• Surveillance Technology

Background:

• Pedestrian detection in infrared surveillance videos is challenging due to varying environmental conditions.
• Existing histogram partitioning methods lack specificity and are sensitive to histogram shapes.

Purpose of the Study:

• To develop a novel candidate generation algorithm for improved pedestrian detection in infrared surveillance videos.
• To enhance background suppression and ensure stable pedestrian presence detection.

Main Methods:

• A novel algorithm combining histogram specification and iterative histogram partitioning.
• Progressive adjustment of dynamic range and efficient background suppression.
• Pre-specification of a uniformly distributed histogram to stabilize histogram shape.

Main Results:

• The proposed method demonstrates improved performance over minimum-cross entropy thresholding.
• Robustness was observed across images acquired under diverse conditions.
• Comparable results were achieved against existing state-of-the-art methods.

Conclusions:

• The novel algorithm effectively addresses limitations of traditional histogram partitioning for pedestrian detection.
• The method provides a stable and reliable approach for identifying pedestrians in infrared surveillance.
• This work contributes to advancements in automated surveillance and safety systems.