This tool takes a video and allows you to put bounding boxes on the frames, without having to do it on individual frames. It will automatically stop when scenes/ camera angels change prompting you for bounding boxes again.
This tool OpenCV's CSRT tracker to achieve this.
python mult_object_tracker.py --name /path/to/video.file which will create the labels in the YOLO format (class_id, relative_center, relative_width, relative height). Currently class_id is simply 0 since it can only look for one class. The images will be stored as a jpg in folder images and corresponding labels in folder named labels.
It will prompt you for bounding boxes in first frame. Draw the box and press enter. You may redraw if you made a mistake as long as you don't press enter. By default you can draw upto 6 boxes. But this can be changed with the --max_obj option.
- If you have less than the maximum number of objects, simply press
enterafter you have drawn the required number of boxes. - If the bounding boxes are of incorrect shapes, press
sto stop the video and redraw. - If you have finished drawing boxes, press
q.
- OpenCV 3+
- Python 3.5+