Efficient last-mile delivery is critical to the optimization of transportation networks, particularly in smart cities that are facing traffic issues. This paper presents a comprehensive approach to optimize delivery processes using Unmanned Aerial Vehicles (UAVs) while reducing environmental impact and resolving logistical challenges. Machine learning models like Convolution Neural Networks (CNNs) particularly You Only Look Once v8 (YOLOv8) obstacle detection model is employed to increase the performance of UAV route optimization. This integration considers taking real-world factors such as battery consumption and dynamic environmental states. YOLOv8 model for obstacle detection along optimal routes can be dealt with proactively through its development. This project optimizes the route which guarantees safety and reliability, in UAV-based deliveries. Our work features the re-routeing procedures which change routes in order to avoid obstacles and this runs smoothly, without delays and it will continue uninterrupted as well. By carrying out empirical assessment and different practical examples, we show efficiency of our method which results in lower travel distances, quicker deliveries and productive overall outcomes. This project work aims to design noticeable transportation systems that are resilient and sustainable for smart cities. It resolves the shortcomings with in the last-mile delivery strategies that currently exist particularly as E-commerce and Q-commerce continue to expanded.