Linear Support Tensor Machine With LSK Channels: Pedestrian Detection in Thermal Infrared Images.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Pedestrian detection in thermal infrared images is difficult due to low resolution and noise.
• Existing methods often struggle with image imperfections and computational efficiency.

Purpose of the Study:

• To propose a robust and efficient pedestrian detection method for thermal infrared imagery.
• To introduce a novel mid-level attribute using local steering kernels (LSK) for improved detection.

Main Methods:

• Utilizing local steering kernels (LSK) as low-level descriptors to capture local image geometry.
• Developing a new image similarity kernel within a support vector machine framework for LSK tensor learning.
• Employing multichannel discrete Fourier transform for accelerating sliding window-based detection.

Main Results:

• The proposed LSK tensor method effectively addresses noise and uncertainty in thermal images.
• The approach facilitates fast and efficient pedestrian localization.
• Experimental validation on public datasets confirms the method's efficacy.

Conclusions:

• The LSK tensor representation offers a powerful tool for pedestrian detection in challenging thermal infrared environments.
• The method provides a significant improvement in both accuracy and speed.
• The release of annotated thermal image data will benefit future research in the field.