Computer vision has been applied to smart cities to address congestion levels in traffic and thereby improve safety within intelligent transportation systems. Congestion results in an increase in traffic flow that, in turn, decreases the speed of the vehicle. Traffic congestion control in a smart city becomes effective only with skillful identification of both pedestrians and vehicles. On the other hand, lightweight UAVs offer a portable and cost-effective option that provides wide-range visibility skills. The challenge, however, comes in the optimization of identification models so that minor and obscured objects are detected. In this text, we utilize the YOLOv11 paradigm. We have pulled out the backbone, YOLOv11, and replaced it with the backbone, Efficient Net. The EfficientNetB7b7 block is used to detail the development of the EfficientNetB7B7-YOLOv11 model to realize vehicle and pedestrian detection using UAVs. Firstly, the neck section of the YOLOv11 adopts the YOLOv11 backbone network in a bid to realize the integration of the EfficientNetB7 to extract global features that are used in identifying the tiny object. And, to tackle the problem of missing object detections, we opted for adopting the EfficientNetB7 in the neck section of the YOLOv11. In this work, we enhance the hybrid between the EfficientNetB7-YOLOv11 to raise the accuracy at which pedestrians and vehicles are detected when they are obscured and improve performance and deal with overlapping bounding boxes. The proposed network improved the performance of object identification by reducing the number of tiny items that were missed. The suggested model was validated to demonstrate its advantages. The effectiveness of the EfficientNetB7 -YOLOv11 model was validated by experimental results, demonstrating a 57.37% improvement in terms of average detection accuracy (mAP0.5) compared to YOLOv11 on the VisDrone2019 dataset.