Target Detection Algorithm for Autonomous Vehicles in Road Scenes

To address the issues of low accuracy and slow processing speeds found in existing road target detection algorithms, as well as the challenges of deploying on embedded devices in practical applications, an improved target detection algorithm based on YOLOv8 (AD-YOLOv8) is proposed. The improvements are achieved through three main techniques: First, the GhostConv module is used to replace traditional convolution modules, effectively compressing the model size while maintaining efficient feature extraction capabilities; second, the integration of the MCA attention mechanism enhances the network’s ability to extract key features and suppress background noise and non-critical information; finally, the CARAFE technology is employed to optimize the up-sampling process, improving attention to target areas in complex environments and spatial information modeling capabilities. Test results on the BDD100K dataset show that the AD-YOLOv8 algorithm significantly increases speed while ensuring high accuracy, achieving a mAP of 62.5%. Moreover, the successful deployment of this model on the Jetson Nano embedded device demonstrates its balance between speed and accuracy, offering a practical and efficient real-time target detection solution for the field of autonomous driving technology.