Simultaneous Localization and Mapping of UAV For Precision Agricultural Application

In the last few years, simultaneous localization and mapping is gaining the popularity in the field of research because of its uses in controlling of robots, unmanned aerial vehicles and autonomous vehicles. The applications of unmanned aerial vehicles are increasing day by day in different fields such as military, agriculture, rescue, and security etc. This needs to localize the UAV that where it is located according to its environment. In this research work, we used a deep learning based visual odometry for pose estimation of unmanned aerial vehicle. This methodology used Convolutional neural network for feature extraction and LSTM is used for learning. We trained and test this model on cotton field dataset for 20 epochs. We calculated the average total, translational and rotation loss for each epoch. Finally, in testing we compute estimated poses and analyze it by graphically with ground truth. Then the results are close to its ground truth and viable.