A YOLO-Based Method for Head Detection in Complex Scenes.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Object detection in complex scenes with dense, occluded objects is challenging.
• Detecting small, randomly distributed objects like heads in intricate environments is particularly difficult for traditional algorithms.

Purpose of the Study:

• To develop a novel head detection algorithm, YOLO-Based Head Detection in Complex Scenes (YOLO-HDCS), for intricate environments.
• To improve the detection of small objects and mitigate overlapping bounding box issues in complex scenes.

Main Methods:

• Introduced two new modules: a context-enhanced, scale-adjusted feature fusion module and an attention-based convolutional module.
• Integrated a modified Intersection over Union (IoU) function to prevent overlapping detections and improve bounding box accuracy.
• Utilized a YOLO-based architecture for head detection in complex scenes.

Main Results:

• Achieved an average accuracy of 82.2% for head detection in complex scenarios.
• Demonstrated significant improvements in multi-scale detection capabilities.
• Showcased rapid loss convergence during the training process.

Conclusions:

• The YOLO-HDCS algorithm effectively addresses the challenges of head detection in complex scenes.
• The developed modules and modified IoU function enhance detection accuracy and efficiency.
• The method offers a robust solution for identifying small, occluded objects in intricate visual data.