An Image Processing-Based Approach for Reading Needle-Type Instruments on Aircraft

Abstract

For guaranteeing the safe and effective functioning of aircraft, image processing techniques can be a valuable tool to detect and evaluate aircraft panel values. In the pursuit of this objective, a dataset covering multiple aircraft models, various sessions, and different lighting conditions was compiled. Four tasks were examined through comparative analysis: object detection, display classification, needle masking, and needle angle detection. YOLOv8 demonstrated high performance in object detection and classification. In the classification task, the adaptability of needle-type device reading was examined by using the well-established models VGG16, Mobilenet V2, and Xception. Denoising autoencoder, U-net, and GrabCut methods were examined for needle masking, and the least squares method was applied to detect needle angle. As we move from the proof-of-concept phase to envisioning the development of an end-to-end system, this work provides significant analysis of image processing methodologies for reading aircraft dashboards.