YOLOv8-Based Distance Estimation for Blind Navigation: Performance Comparison of OpenCV and Coordinate Attention Techniques

Blindness presents a significant challenge in the development of assistive technologies, particularly for navigation, as it requires accurate distance perception to enable effective mobility for the visually impaired. The research addresses this issue by evaluating and comparing the performance of the YOLOv8 model integrated with OpenCV and the Coordinate Attention Weighting (CAW) technique for distance estimation in blind navigation systems. The main research objective is to improve distance estimation accuracy without the need for additional sensors. Initially, YOLOv8 with OpenCV shows less optimal results, prompting efforts to enhance its performance to surpass the effectiveness of CAW, while maintaining a sensor-free solution. The research then integrates YOLOv8 with OpenCV for baseline comparison and applies CAW for advanced feature attention in the distance estimation process. The research also integrates mathematical formulations for camera calibration and depth estimation, utilizing techniques such as triangulation and reprojection to refine the accuracy of object distance prediction. The results show that improved YOLOv8 + OpenCV significantly outperforms original YOLOv8 + OpenCV, with reduced Mean Squared Errors (MSE) across various distance intervals (0-1 m, 1-2 m, 2-3 m, 3-4 m, and 4-5 m). YOLOv8 + CAW also shows improvement compared to the original YOLOv8 + OpenCV but does not surpass the performance of the improved OpenCV integration. These findings demonstrate the potential of refined computer vision techniques in achieving high-accuracy and sensorfree distance estimation, enhancing real-time navigation systems for the blind. The research paves the way for further advancements in the development of accessible and reliable navigation technologies for the visually impaired.