Detecting and Counting humans in Aerial Image ( Drone Vision )

• Usecase implemented using YOLOv3 ( Real-Time Object Detector )
• YOLOv3 Detects images with a speed of 45 FPS, It works faster when compared with other Real-Time Detectors like R-CNN and Faster R-CNN etc
• Please find Notebook attached in this Repo for Implementation Details

• Total Humans: 6

• Model trained on COCO dataset contains 80 classes including humans ( needed for our usecase )
• Because of lack of dataset related to our usecase, Custom training is not possible. So, I have decided to use pretrained model ( trained on COCO dataset ) to detect humans in the given Image
• Model couldn't able to detect too small humans present in the image, this could be resolved by using Sliding window as a part of the pipeline

• By Implementing Sliding Window would improve detection accuracy even if we are using Pretrained Model

• Custom training the Model with our own dataset will improve Accuracy
• Sliding Window helps Model to identify tiny objects in Low Resolution Images