Pedestrian Counting with Occlusion Handling Using Stereo Thermal Cameras

Area of Science:

• Computer Vision
• Robotics
• Sensor Fusion

Background:

• Accurate pedestrian counting is vital for urban planning, retail, and event management.
• Traditional visual cameras struggle with variable lighting and occlusions, limiting day-and-night effectiveness.
• Existing automated counting methods face challenges in complex, crowded environments.

Purpose of the Study:

• To develop and evaluate a robust pedestrian counting system using stereo thermal cameras.
• To address limitations of visual sensors in varying light and occlusion scenarios.
• To propose a 3D point cloud-based algorithm for accurate people counting.

Main Methods:

• Utilized a stereo thermal camera setup for 3D point reconstruction.
• Developed a novel algorithm for pedestrian counting based on clustering and tracking of 3D point clouds.
• Tested the system on video sequences from a public event with moderate pedestrian density and occlusions.

Main Results:

• The stereo thermal camera system demonstrated high accuracy in pedestrian counting.
• Success rates of 95.4% and 99.1% were achieved on two challenging video sequences.
• The proposed 3D point cloud algorithm effectively handled occlusions and varying conditions.

Conclusions:

• Stereo thermal imaging offers a viable solution for reliable, all-day pedestrian counting.
• The developed 3D point cloud processing algorithm significantly improves counting accuracy in complex scenes.
• This technology has strong potential for applications in smart city initiatives and crowd management.