An Improved YOLOv8 Algorithm for Icon Detection in Vehicle Dashboard

Icon detection in vehicle dashboard is a crucial step in the automated testing of vehicular systems. However, the lack of publicly available datasets for the vehicle dashboard, Automated testing has high real-time detection requirements, combined with existing YOLO (You Only Look Once) models exhibit limitations in detecting small targets effectively, and In practical engineering applications, using camera to capture the vehicle dashboard for icon detection may lead to issues such as strong reflections from ambient light sources, resulting in bright strip and striping. These phenomena hinder existing algorithms’ ability to achieve rapid and accurate detection of vehicle dashboard icons. To address these challenges, this paper constructs a new dataset specifically for car dashboard icon detection, including images of automobile dashboards captured under conditions with strong reflections, bright strips, and stripes, which are challenges commonly encountered in real engineering applications, thus filling the gap created by the lack of suitable existing datasets. Additionally, this work enhances the existing YOLOv8 model by proposing the YOLOv8-ABC feature extraction algorithm, which improves the network’s capability to detect display icons by clustering all anchors in the training set with an improved K-means++ algorithm. This approach yields 12 anchors tailored to the in-vehicle system icon dataset, replacing the anchor-free module in the original model. This approach reduces the number of parameters in the model, which reduces the computational complexity, making the model more suitable for practical engineering applications. Meanwhile, by incorporating BiFPN (Bi-directional Feature Pyramid Network) and CBAM (Convolutional Block Attention Module) into the model, the efficiency of the model is ensured while significantly improving the detection accuracy and improves the detection ability of the model for small targets. Experimental results demonstrate that, compared to the original YOLOv8 model, the YOLOv8-ABC algorithm proposed in this paper improves mAP50 and mAP50-95 by 2.8% and 2.6%, respectively, while significantly reducing the number of parameters required by the model. These improvements indicate that the improved algorithm is able to detect icons in vehicle dashboards in real-time with improved detection accuracy, and thus can be effectively applied to automated testing of vehicle dashboards.