The Idea | Technical Overview | Setup | With Docker

Open source app for tracking auditory boredom on video

In the current project we develop a framework for video analysis of public presentations and events. Given a video, this tool tracks viewers attention during lectures or presentations.

With this app you can assess speaker’s skills or detect the most resonant parts of a presentation, or give a summary of some socio-demographic statistics such as age and gender distribution without conducting surveys.

Slides from presentation of our project (russian)

People tracking example video

Abstract

The boremeter pipeline includes four main components:

• Faces detection
• Faces tracking
• Faces recognition
• Generating report & visualising results

For faces detection we use fast and robust Viola-Jones algorithm.

More on Viola-Jones algorithm you can read in original paper

Faces tracking is implemented using sparse optical flow and Lucas–Kanade method.

You can find out more in Lucas-Kanade in a Nutshell

On the recognition stage we use Caffe framework and pre-trained convolutional neural networks from to exract age and gender from cropped images of detected faces. The pre-trained models are taken from http://howhot.io project.

The output of the pipeline is a rendered HTML file which contains:

• graphs of percentage of people interested in the presentation over time
• gender representation
• age representation

Boremeter can also visualize detection results and output an .avi video file with detected faces.

• Ubuntu 14.04 or older (or OS X)
• Python 2.7
• caffe
• OpenCV3

Download pre-trained caffe nets and save them locally to {caffe_root}/models:

• age.prototxt
• age.caffemodel
• gender.proto
• gender.caffemodel

To install boremeter just run

$ git clone https://github.com/walterdd/Boremeter.git
$ cd Boremeter
$ python setup.py install

boremeter --file={input video file path}

To find out more about boremeter usage

boremeter --help

We strongly suggest you to run boremeter in Docker. That will make life easier.

To install boremeter just clone the repository and build with Docker

$ git clone https://github.com/walterdd/Boremeter.git
$ cd Boremeter
$ docker build -t boremeter .

To run Boremeter in Docker use

$ docker run -v {host directory path}:{container directory path} -it boremeter
$ boremeter --file={input video file}