A Yolov8-Based Object Detection Framework with Moiré Pattern Removal

Intelligent vehicle testing platforms, based on vehicle-in-the-loop and digital twin technologies, have emerged as critical simulation tools. These platforms integrate sensors, computing units, and cameras within physical infrastructures while generating synthetic traffic data via front-end displays. However, camera-captured images often exhibit moiré interference when imaging display screens, significantly degrading the performance of conventional object detection algorithms. To address this challenge, we propose an enhanced detection framework termed Moving Moiré Removal and Object Detection Network (MMROD-Net). In this framework, we introduce a dynamic Frame Extraction and Crop-Resize (FECR) module for real-time moiré suppression. To strengthen feature representation, an Efficient Multi-Scale Attention (EMA) mechanism is integrated into the Neck network of Yolov8. Besides, we adopt a pseudo-label based incremental learning strategy for iterative model refinement. Experimental evaluations on the MMROD dataset demonstrate that MMROD-Net achieves a 34% improvement in mean Average Precision (mAP) compared to baseline Yolov8 networks, validating its efficacy in addressing moiré-induced detection degradation.