PC-CS-YOLO: High-Precision Obstacle Detection for Visually Impaired Safety

Area of Science:

• Computer Vision
• Artificial Intelligence
• Assistive Technology

Background:

• Obstacle avoidance and safety are critical for visually impaired individuals.
• Complex street environments present significant challenges for current blind obstacle detection systems.
• Existing solutions often lack real-time accuracy for effective obstacle avoidance.

Purpose of the Study:

• To develop an advanced blind obstacle detection system.
• To improve the real-time performance and accuracy of obstacle avoidance for the visually impaired.
• To address the limitations of current systems in complex urban settings.

Main Methods:

• Proposed a novel blind obstacle detection system utilizing the PC-CS-YOLO model.
• Enhanced the backbone network with a partial convolutional feed-forward network (PCFN) to minimize computational load.
• Introduced a Cross-Scale Attention Fusion (CSAF) mechanism for robust multi-scale feature integration across sensory domains.

Main Results:

• Achieved improvements of 2.0% in precision, 3.9% in recall, and 1.5% in mAP50 compared to state-of-the-art networks.
• Demonstrated a GPU inference speed of 20.6 ms, which is 15.3 ms faster than YOLO11.
• The system meets real-time requirements for effective blind obstacle avoidance.

Conclusions:

• The proposed PC-CS-YOLO based system significantly enhances blind obstacle detection capabilities.
• The integration of PCFN and CSAF contributes to improved accuracy and efficiency.
• This system offers a promising solution for real-time, reliable obstacle avoidance, improving safety for visually impaired individuals.