Navigation Assistive Intelligent Device with YOLOv8 Object Detection and Geometric Distance Estimation

Objectives: This research aims at developing a real-time wearable navigation system to guide the visually impaired users, capable of detecting, classifying, tracking, and localizing obstacles with the help of YOLOv8, Deep SORT, and geometrical distance-angle estimation. It is aimed at providing a safe, precise, and context-driven navigation aid with an embedded portable platform. Methods: The system combines an embedded Raspberry Pi camera and a YOLOv8 object detector, Deep SORT multi-object tracker, and a geometric model of pinhole cameras to estimate distance and angular position. Video frames are processed in real-time and sent through the detection tracking pipeline and then a safety labeler is used to provide the obstacles with a SAFE or DANGER label. Accuracy, precision, recall, F1-score, normalized confusion matrix, and Mean Absolute Error (MAE) of distance and angle estimation are all performance evaluation metrics. Findings: Under realistic conditions, experimental results indicate that the reliability is high with a 95% 100% accuracy, 92% 98% precision, 88% 95% recall and 90% 96% F1-score. There is low MAE in distance estimation and the error in angular estimation is within safe limits in navigation. According to the confusion matrix analysis, the classification accuracy is high in the case of navigation-critical objects (e.g., person, door, TV), and the confusion is low in the case of similar classes in appearance (chair–couch). All in all, the system is very stable in FPS and has a constant detection confidence of dynamic environments. Novelty: The proposed system is the only one to integrate lightweight YOLOv8 detection, Deep SORT tracking and camera-based geometric distance-angle estimation into one wearable assistive system, unlike the current system which only uses detection or depth sensors. Besides, it combines a real-time safety-level classifier (SAFE/DANGER) and improves the situational awareness, which makes the system more viable in the case of navigation by the blind. It is an integrated, embedded, inexpensive architecture that provides rapid, precise, and reliable environmental awareness that can be used in real-time mobility support. Keywords: YOLOv8, Deep SORT, Distance angle estimation, Assistive technology, Safety level classification, Computer vision