Addressing the concerns stemming from the complexity of parameters and significant computational demands encountered in contemporary vehicle object detection algorithms tailored for traffic surveillance applications, which hinder efficient deployment on devices with limited resources, we introduce GSE-YOLOv8, a streamlined variant of YOLOv8 tailored specifically for this purpose. Our innovation leverages Ghost convolution advancements to conceive C2fGhostv2, an optimized feature extraction module that achieves a reduction in computational footprint while preserving robust feature capturing abilities. To further enhance performance, we integrate the Shuffle Attention (SA) mechanism within the neck network architecture. This enables dynamic prioritization of feature map weights, emphasizing salient characteristics and minimizing distracting background influences during vehicle detection. Additionally, we refine the YOLOv8n model by introducing the MPDIOU loss function, which not only accelerates convergence but also appreciably boosts segmentation precision and model generalizability when compared to the baseline loss function. Evaluations conducted on the UA-DETRAC traffic dataset validate the efficacy of GSE-YOLOv8, exhibiting an enhancement in detection accuracy sans compromise on precision. Notably, our approach achieves a substantial reduction of 32.30% in parameters and 30.33% in computational requirements vis-à-vis the original YOLOv8, underlining its practical relevance and feasibility for real-world implementations.