Robust Pedestrian Classification Based on Hierarchical Kernel Sparse Representation

Area of Science:

• Computer Vision
• Autonomous Systems

Background:

• Pedestrian detection is crucial for autonomous vehicles to ensure safety.
• Challenges include varying illumination, complex backgrounds, and occlusions in outdoor environments.

Purpose of the Study:

• To propose a novel hierarchical feature extraction and weighted kernel sparse representation model for robust pedestrian classification.
• To enhance the performance of pedestrian detection systems in autonomous vehicles.

Main Methods:

• Hierarchical feature extraction using CENTRIST descriptor and max pooling for appearance invariance.
• Kernel sparse representation model with a Gaussian weight function to handle occlusions.

Main Results:

• The proposed method demonstrates robust performance on benchmark datasets (INRIA, Daimler) and real-world occluded datasets.
• Achieved superior pedestrian classification compared to existing state-of-the-art methods.

Conclusions:

• The novel model effectively addresses challenges in outdoor pedestrian detection.
• The method offers a more robust solution for autonomous vehicle perception systems.