Intelligent Vehicle Object Detection Based on Improved YOLOv7 Algorithm

Abstract Intelligent driving of automobiles is based on environment awareness, and the technology that makes environment perception possible is target detection. Road traffic scenarios are complicated and variable, which can lead to issues including misdetection and omission of small objects and obscured areas. Furthermore, because to their limited processing capacity, intelligent cars must not only have a high detection accuracy and speed but also a low-complexity model in order to lessen their reliance on high-performance computing platforms. When compared to other lightweight models, the lightweight version of the YOLOv7 algorithm offers a lot of opportunity for improvement in terms of processing and parameter count. Thus, based on the enhanced YOLOv7 algorithm, this research suggests an intelligent vehicle target recognition technique. The YOLOv7-tiny network structure serves as the foundation for the addition of a small target detection layer, the introduction of a context aggregation module to enhance the feature extraction process and the model's capacity to detect small targets in complex scenarios, the reconfiguration of the SPPCSPC module to speed up detection, and the proposal of the P-ELEN module, which reduces the model's number of parameters and computation load, to lessen redundant computation and memory access. Nine sets of comparison tests, eight sets of ablation experiments, and detection effects are used to validate the enhanced YOLOv7 algorithm's detection performance. The experimental findings demonstrate that the enhanced YOLOv7 algorithm increases detection accuracy while simultaneously decreasing model complexity.