Perceptual fusion-based target detection model for YOLOv8 autopilot

In scenarios such as automatic driving and industrial monitoring, which require high real-time and accuracy, traditional target detection matches the images using pre-prepared target templates, but these target scale change, rotation and other problems are poorly adapted. In view of this, this paper proposes a perceptual fusion YOLOv8 model design based on perceptual fusion, which achieves perceptual fusion of visual data with multi-source heterogeneous sensor data, such as infrared thermal imagery, in the spatial channel two-dimensionality by constructing a cross-modal feature fusion mechanism. Joint training with multimodal data in the training phase increases the generalisation capability of the model, allowing it to adapt to many different sensor combinations and changing environmental changes. In this way, YOLOv8 implements a real-time target detection function by dividing the visual image into a grid and predicting multiple bounding boxes simultaneously, which is suitable for application scenarios where autonomous driving is reactive. The proposed model in this paper is validated on two benchmark datasets used for autonomous driving research, and the target detection method in this paper using the YOLOv8 network in the KITTI dataset is more effective and robust.