The proposed approach focuses on autonomous UAV (Unmanned Aerial Vehicle) navigation and obstacle detection using only visual sensors specifically, the on-board front camera. Unlike traditional methods that rely on multiple sensors (e.g., LiDAR, radar, or GPS), this vision-based system aims to reduce hardware complexity and cost while maintaining robust performance in diverse environments. By leveraging computer vision and deep learning techniques, the UAV processes real-time camera feed data to detect obstacles, map surroundings, and plan collision-free paths. Key challenges include handling dynamic environments, varying lighting conditions, and real-time processing constraints. The system employs feature extraction, depth estimation, and semantic segmentation to interpret visual data, enabling the UAV to navigate autonomously without external aids. Advantages of this approach include reduced sensor dependency, lower power consumption, and improved adaptability in GPS-denied or cluttered spaces (e.g., indoor settings or dense urban areas). However, limitations may arise in low-visibility conditions (e.g., fog or darkness) or with texture-less surfaces that complicate depth perception. The method aligns with advancements in lightweight AI models optimized for edge computing, ensuring efficient onboard processing. Future enhancements could integrate multi-camera setups or hybrid sensor fusion for increased reliability. Overall, this vision only navigation strategy offers a scalable and cost-effective solution for UAV autonomy, particularly in applications like surveillance, inspection, and disaster response where simplicity and agility are critical.