In modern transportation systems, applying effective pedestrian detection techniques to the truck driving perspective is crucial for improving traffic safety, however, traditional computer vision methods often perform poorly when dealing with occluded, long-distance, and pedestrians with varying postures. To solve these problems, this paper proposes a pedestrian detection approach based on a multi-strategy image recognition improvement mechanism. First, based on the YOLOv8 algorithm, Mathematically Proven Distance-IoU (MPDIoU), Complete Intersection over Union Loss (CIoU Loss), and Distributed Focus Loss (DFL) strategies. Then, Dynamic Snake Convolution (DSC) and Deformable Convolution (DC) are designed to solve the problems of different pedestrian postures and pedestrian misdetection and omission. At the same time, we construct the Occlusion Sensing Attention Mechanisms, Context Aggregation Attention Mechanisms (CAAMs), and other mechanisms to solve the problem of pedestrian occlusion and recognition accuracy in complex scenes. Finally, the performance of the proposed method is verified based on the real-world dataset, and compared with the YOLOv8 algorithm, the technique improves by about 80% in terms of false detection rate improvement, 36.84% in terms of leakage detection rate improvement, 92% in terms of recognition frame accuracy improvement, and 95.65% in terms of recognition frame overlap improvement. The method in this paper significantly improves several important performance indicators, which strongly highlights the excellent advantages of the technique.