<p>With the rapid development of intelligent transportation systems, the detection of vehicles and pedestrians has demonstrated significant application value in the field of autonomous driving. However, target detection in complex road environments faces numerous challenges, including target occlusions, high resource demands of algorithms, and the varying scales of vehicles and pedestrians. Given the limitations of the YOLO series algorithms in detecting multi-scale targets in complex road scenarios, we propose a method named LMF-YOLO to effectively address these challenges. Firstly, we designed a feature extraction module based on a star structure with a large separable kernel (C2f-SLSK) to enhance the extraction of both local and global information. Secondly, we introduced a new Context and Local Feature Fusion (CLFF) module, which adaptively fuses the extracted multi-scale target feature information through weighted feature fusion. Thirdly, we designed a new Mixed Information Attention (MIA) mechanism to adaptively integrate global and local information, as well as channel and spatial information. Finally, in conjunction with MIA, we developed a feature pyramid network that fuses high-level and low-level features (HLFPN). This design not only improves the representation capability for multi-scale target features but also makes the model more lightweight. Experimental results show that compared with other state-of-the-art models, our LMF-YOLO method demonstrates superior performance and effectiveness, making a significant contribution to the field of road target detection in autonomous driving.</p>