This project contains three main modules:
1- Faster RCNN module is executed every certain number of frames and a number of vehicles are identified (the first frame is analysed by this module).
2- Optical flow is executed and a number of boxes which may contain a vehicle are identified. This is run every certain amount of frames but much smaller than in the case of the Faster RCNN module
3- CNN module checks the boxes identified by the optical flow module and decides if the box contains a vehicle or not. This is executed for every box identified by the optical flow module.
The result of this project is the amount of cars that appear in each video.
Pull docker image from:
https://hub.docker.com/r/squintana/dl-docker/
Download the full project from the following location:
https://drive.google.com/open?id=0B4RgtXiS2li0QksxaUFpRWxBaEE
Place the downloaded project in the docker sharedfolder (<local_folder>).
Run docker image using the following command:
nvidia-docker run -it -p 8888:8888 -v ~/<local_folder>:/root/sharedfolder squintana/dl-docker bash
Run ~/sharedfolder/tools/main.py to obtain the results.
Videos are stored in tf-faster-rcnn-master/tools/videos_test1
- Python 2.7.x
- TensorFlow == 1.0.1
- OpenCV 3.3
- Cuda