UVid-Net: Enhanced Semantic Segmentation of UAV Aerial Videos by Embedding Temporal Information

Semantic segmentation of aerial videos has been extensively used for decision making in monitoring environmental changes, urban planning, and disaster management. The reliability these support systems is dependent on the accuracy video semantic algorithms. existing CNN-based methods have enhanced image by incorporating an additional module such as LSTM or optical flow computing temporal dynamics which a computational overhead. proposed research work modifies CNN architecture information to improve efficiency segmentation. In this work, encoder-decoder based (UVid-Net) unmanned vehicle (UAV) encoder embeds temporally consistent labeling. decoder introducing feature-refiner module, aids accurate localization class labels. UVid-Net UAV quantitatively evaluated extended ManipalUAVid dataset. performance metric mean Intersection over Union 0.79 observed significantly greater than other state-of-the-art Further, produced promising results even pretrained model street scene fine tuning final layer videos.