TRACKING AIDED DRONE BIRD CLASSIFICATION USING YOLO AND LSTM

In this work, we propose a system that uses thermal imaging and deep learning techniques to detect and classify objects in real-time. We use thermal images provided in the ICIP challenge dataset, which are then fed into a (You Look Only Once) YOLOv5 object detection model to detect and locate objects in the scene. The detected objects are then tracked using a centroid tracking algorithm to generate trajectories of the objects. We then use the trajectory data to train a Long Short Term Memory (LSTM) model for object classification based on the object’s movement pattern. We demonstrate the effectiveness of our system on a dataset of thermal images captured in a variety of scenarios. The results show that our system can accurately detect and classify objects in real-time, even in challenging conditions such as low-light environments. We have also used ResNet-18 model to identify the different types of distortions in the dataset. Our approach has potential applications in surveillance, security, and search and rescue operations.