I have successfully trained custom weights to detect pedestrians/Cars/Busses/motors...etc from drone video clips!
Go ahead and download the weights I have trained from: https://www.dropbox.com/s/ashde9hjwfefm4i/yolo-obj_best.weights?dl=0
Paste these weights into the backup folder under darknet
After this make open the darknet_video.py and set the input video path.....Make sure it is a drone footage. See the results and enjoy.
You might want to first make the darknet folder. Make the necessary changes in the make file,and execute make
For the dataset I used VisDrone's Data to train the model.
https://www.dropbox.com/s/w0t148ig7us67es/res.avi?dl=0 The above link is the result of how detections are happening in videos