Facial Emotion Recognition in a video stream using a pre-trained Deep-CNN model.

The video stream is resized and limited to about 5FPS to boost performance and reduce delays.

The model used was trained using this repository: https://github.com/obensch/fer-2013

Python version: 3.7

The following python packages are required

• numpy
• openCV
• imutils
• time
• tensorflow/keras

Webcam is enabled by default in all scripts. To enable webstream comment the line:

cam = cv2.VideoCapture(0)

and uncomment the line

# cam = cv2.VideoCapture("http://192.168.178.21:8080/video?type=some.mjpg")

To change the frame settings edit the folling lines in all the scritps:

# set frame size
FrameWidth = 1280
FrameHeight = 720

Requirements: The files 'haarcascade_frontalface_default.xml' is required for face detection. The file can be downloaded e.g. from here: https://github.com/opencv/opencv/tree/master/data/haarcascades

This script can be executed using the command:

python main.py

The model can be changed in line 19:

model.load_weights('main.h5')