Nighttime Foreground Pedestrian Detection Based on Three-Dimensional Voxel Surface Model

Area of Science:

• Computer Vision
• Surveillance Technology
• Artificial Intelligence

Background:

• Pedestrian detection is crucial for surveillance but challenging at night.
• Existing methods struggle with low illumination and occlusion.
• Expensive thermal cameras are often required for effective nighttime detection.

Purpose of the Study:

• To develop an affordable and effective solution for nighttime pedestrian detection.
• To introduce a novel 3D foreground pedestrian detection model using near-infrared stereo vision.
• To enable real-time segmentation and location of pedestrians in low-light conditions.

Main Methods:

• Utilized a near-infrared stereo vision system with calibrated network cameras and lamps.
• Developed a novel voxel surface model for estimating 3D geometric changes.
• Implemented a free update policy for unknown points and shadow extraction for false alarm removal.

Main Results:

• The system successfully performs real-time nighttime pedestrian segmentation and detection, even under heavy occlusion.
• Outperformed classical background subtraction and RGB-D methods.
• Achieved performance comparable to state-of-the-art deep learning methods at a significantly lower hardware cost.

Conclusions:

• The proposed near-infrared stereo vision system offers an affordable and effective solution for nighttime pedestrian detection.
• The novel 3D voxel surface model demonstrates robust performance in challenging low-light and occluded environments.
• This approach provides a viable alternative to expensive thermal imaging systems for surveillance applications.