Skip to content

Latest commit

 

History

History
49 lines (44 loc) · 1.3 KB

README.md

File metadata and controls

49 lines (44 loc) · 1.3 KB
Raw

Introduction

robot control via eye action based on cnn+lstm

* step1: camera frame -- face detection(Haar+Adaboost)
* step2: face region(step1) -- eyes detection(left+right, Haar+Adaboost)
* step3: eyes region(step2) -- eyes selection(lenet-5)
* step4: eyes region(frame difference or optical flow or original pic) -- cnn+lstm classification
* step5: predict -- action control(finite state machine)

Dependences

  • numpy, scipy
sudo apt-get install python-dev, python-numpy, python-scipy
  • skimage
sudo apt-get install python-skimage  or  pip install scikit-image
  • sklearn
sudo apt-get install python-sklearn  or  pip install scikit-learn
  • opencv
sudo apt-get install python-opencv

Demo/Usage

Before run demo, you should train network with your own data, at first prepare your data. In file train.py, modify the param

face_cascade_file='data/model/haarcascade_frontalface_alt.xml',
eyes_cascade_file='data/model/haarcascade_eye.xml'
left_video_dir = 'data/video/left',
right_video_dir = 'data/video/right',
recover_video_path = 'data/video/recover/IMG_1950.MP4'

And then run the script train.py to train the model

python train.py

At last, please run demo

python demo.py