UAV object detection has to be very accurate and extremely fast to process the live feed and give instant predictions of the current terrain. Our project uses a very optimal object detection algorithm - YOLO to accurately map the environment and detect various classes of objects. We will optimize the algorithm such that we can use it via live camera feed of UAV and it will be able to classify the environment precisely. Our main objective is to use Deep Learning to make predictions on its own where humans cannot intervene or where there isn’t enough visibility.
We are optimizing YOLOv3 algorithm which is much faster and accurate than R-CNN and other object detection algorithms. We will use our custom dataset to train and model the data which can be used in real world scenarios.
-
Search and Rescue during Natural disaster: In Tsunami or earthquake conditions rather than manually searching for people and animals, our algorithm will detect it on its own with high accuracy.
-
Pothole detection: It can be used by Government agency to check the current conditions of the road. Thus it saves human time and can reach places much faster where humans cannot.
-
Bird obstacle detection: The main problems that UAVs face is being damaged by incoming birds in flight. Our model will prevent such scenarios and divert the UAV to a different projectile.
-
Traffic and accident monitoring: In cases of emergency where road is crowded/blocked, the drones will be be able to access the exit points on the own and give live telemetry information to base.
-
Forest Fire detection: The trained model will detect the forest fire source and can predict the possible.
