Real-Time-Object-Detection_YOLO

Real Time Object Detection using Darknet YOLO (You Only Look Once) algorithm, OpenCV on the COCO (Common Objects in Context) dataset.

Implementation

• Darknet YOLOv4 has been used to along with the YOLOv4 preconfigured weights to the sameple clip. The output clip is recieved at around 8 FPS (on NVIDIA GTX 1050Ti GPU).
  
• You may use the files in the following link for YOLOv4 weights, cfg, notebook etc. Link: https://drive.google.com/drive/folders/1rl4nKHJ9WrGyAS2tumx3lxMYb1i1ziSe?usp=sharing

Increasing FPS

• Operations like fetching video from web camera are I/O intensive, we make use of threading (imutil.py) to make use of seperate thread for I/O related operations and end up in a heavily increased frame rate.

Output

The output video is generated through OpenCV's VideoWriter to output_clip.mp4.

Watch the video here: