Augmentation for Accuracy Improvement of YOLOv8 in Blind Navigation System

Abstract

This study addresses the critical need for enhanced accuracy in YOLOv8 models designed for visually impaired navigation systems. Existing models often struggle with consistency in object detection and distance estimation under varying environmental conditions, leading to potential safety risks. To overcome these challenges, this research implements a rigorous approach combining data augmentation and meticulous model optimization techniques. The process begins with the meticulous collection of a diverse dataset, essential for training a robust model. Subsequent preprocessing of images in the HSV color space ensures standardized input features, crucial for consistency in model training. Augmentation techniques are then applied to enrich the dataset, enhancing model generalization and robustness. The YOLOv8 model is trained using this augmented dataset, leading to significant enhancements in key performance metrics. Specifically, mean average precision (mAP) improved by 13.3%, from 0.75 to 0.85, precision increased by 10%, from 0.80 to 0.88, and recall rose by 10.3%, from 0.78 to 0.86. Further optimization efforts, including parameter tuning and the strategic integration of a Kalman Filter, notably improved object tracking and distance estimation capabilities. Final validation in real-world scenarios confirms the efficacy of the optimized model, demonstrating its readiness for practical deployment. This comprehensive approach showcases tangible advances in navigational assistance technology, significantly improving safety and reliability for visually impaired users.

Keywords

• yolov8
• object detection
• data augmentation
• model optimization
• visual impairment