Research on Object Detection in Cluttered Hospital Corridor Scenes with CSAWOA-YOLOv8.

Area of Science:

• Computer Vision
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Hospital corridors present complex environments with challenges like varying object scales, occlusions, and dense small objects.
• Existing object detection methods struggle with accuracy and efficiency on resource-constrained platforms in these dynamic settings.

Purpose of the Study:

• To develop a high-precision object detection framework, CSAWOA-YOLOv8, specifically for complex medical corridor environments.
• To enhance feature representation and address challenges such as target occlusion and small-object recognition.

Main Methods:

• Proposed a CSAWOA-YOLOv8 model integrating semantic and low-level features (texture, color) for discriminative representation.
• Introduced the T-CBS module for shallow feature extraction and global context integration to handle occlusions.
• Incorporated the BiFormer module to improve feature discriminability and small-target recognition.
• Modified the classification function to address class imbalance and developed CSAWOA for hyperparameter optimization.

Main Results:

• Achieved improvements of 4.9% in mAP, 6.1% in precision, and 8.3% in recall compared to YOLOv8.
• Demonstrated enhanced generalization capabilities and robustness against background noise.
• The CSAWOA optimization effectively balanced detection diversity and convergence speed.

Conclusions:

• The CSAWOA-YOLOv8 framework offers a reliable and efficient solution for object detection in complex hospital corridors.
• The proposed methods provide a strong foundation for advancing real-world healthcare applications and future research in medical imaging.
• This approach effectively tackles key challenges in dynamic medical environments, improving detection performance significantly.