BEST RESULT :
Model type : 3D CNN + FC [base - VGG19]
Accuracy : 68.54%
Current structure :
object detection -> Mask the background -> frame diff -> 2D CNNlstm -> FC1 -> output
Links
Models link : https://github.com/OlafenwaMoses/ImageAI/releases/tag/1.0/
imageai : https://imageai.readthedocs.io/en/latest/
Future
YOLOV4 : https://www.youtube.com/watch?v=iPwepy-SVCQ