Pedestrian Detection Using Integrated Aggregate Channel Features and Multitask Cascaded Convolutional...

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Pedestrian detection is crucial for autonomous systems but is challenged by diverse human appearances.
• Deep learning detectors require extensive datasets and high-performance GPUs, limiting their application in resource-constrained scenarios.

Purpose of the Study:

• To propose a novel pedestrian detection approach that overcomes limitations of deep learning methods.
• To develop a detector suitable for scenarios with limited datasets and computational resources.

Main Methods:

• Integration of a pretrained multitask cascaded convolutional neural networks face detector with a traditional aggregate channel features pedestrian detector.
• Utilizing a score combination module to merge features from both detectors.
• Comprehensive parameter optimization and cross-dataset validation on INRIA, ETHZ, Caltech, and Citypersons datasets.

Main Results:

• The integrated detector significantly increases recall and decreases the log-average miss rate compared to using the traditional detector alone.
• Achieves performance comparable to Faster R-CNN on the INRIA test set, outperforming the standalone Aggregated Channel Features detector.
• Demonstrates robustness across different training sets, test sets, and thresholds.

Conclusions:

• The proposed hybrid detector offers a promising solution for pedestrian detection with limited data and computational power.
• This method provides a practical alternative to solely relying on computationally intensive deep learning models.