Depth Analysis for Unmanned Aerial Vehicle using Incremental Deep Learning Approach and Publisher-Subscriber Model
Sakshi Balnath Bankar, Sandhya Mathur, N. Pathak, Aishwarya Girish Gawas, Dipti D. Patil · 2024
Depth estimation is crucial for unmanned aerial vehicles (UAVs) because it enables obstacle avoidance, terrain mapping, precision landing, object tracking, and autonomous navigation. This paper presents an innovative integration of the single board computation device (like NVIDIA Jetson Nano platform), a deep-learning incremental model that is deployed for depth estimation, facilitating enhanced navigation of UAVs. Through the incorporation of incremental learning, the model adapts to changing environmental conditions, thereby improving the accuracy of depth estimation. Furthermore, consolidating Message Queuing Telemetry Transport (MQTT) protocol displaying the captured image and estimated depth values on a mobile application enables seamless interaction with the UAV system. This collaborative framework between Jetson Nano and mobile visualization establishes a robust foundation for streamlined UAV operations.