Title: Enhancing Real-Time Automatic License Plate Recognition via YOLOv8, SORT Tracking, and Temporal Data Interpolation

Abstract:

The computational demands of real-time video processing often restrict the deployment of Automatic License Plate Recognition (ALPR) systems, particularly within dynamic traffic monitoring environments. Achieving high-accuracy recognition under unconstrained conditions—such as extreme lighting shifts, rapid camera movements, high vehicle velocities, and physical obstructions—remains a significant challenge. These variables frequently result in fragmented tracking trajectories and suboptimal Optical Character Recognition (OCR) performance. To address these limitations, this study introduces a comprehensive five-stage, end-to-end algorithmic pipeline. This framework ensures a seamless integration of deep learning-based object detection, kinematic multi-object tracking, and geometric temporal data interpolation.

The proposed architecture leverages the computational efficiency of the YOLOv8 nano model for initial vehicle localization. Subsequently, the Simple Online and Realtime Tracking (SORT) algorithm is employed to establish spatial-temporal connections across video frames. For character recognition, a specialized variant of the YOLOv8 detector isolates license plate regions; the extracted image segments are then processed by EasyOCR, subject to positional syntax verification constraints. Crucially, the system incorporates an offline temporal interpolation mechanism for bounding boxes, designed to reconstruct and stabilize previously disjointed tracking paths.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC