Skip to content

This repository contains the source code for the paper Motion and Region Aware Adversarial Learning for Fall Detection with Thermal Imaging

License

Notifications You must be signed in to change notification settings

ivineetm007/Fall-detection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Motion and Region Aware Adversarial Learning for Fall Detection with Thermal Imaging

This repository contains the source code for the paper Motion and Region Aware Adversarial Learning for Fall Detection with Thermal Imaging by Vineet Mehta, Abhinav Dhall, Sujata Pal and Shehroz S. Khan.

In this work, we formulate fall detection as an anomaly detection within an adversarial framework using thermal imaging. Fall events are detected by using spatio-temporal autoencoders trained in an adversarial manner. A fall event is associated with the person and its interactions with nearby objects. We extract the region around the person by tracking the person in the video using a pre-trained object detector and also extract the optical flow in the region for motion-based discriminative learning.

Examples

Person Tracking

We use an object detector along with Kalman filters to track the bounding box for the person. We apply contour box localization to improve the overall tracking. It consists of finding the smallest bounding box that surrounds the biggest white blob in a thermal image. The below animation shows the effect of tracking with contour box localization. The left one is generated by Kalman filtering with an object detector, and the right is along with the contour box.

alt="tracking
Left- Tracking without Contour Box, Right- Tracking with Contour Box

Dark Blue- Detection Box, Sky Blue- Tracker Prediction Box, Green - Contour Box (If Matched with other boxes), Red - Otsu Contour (Not matched) and Yellow- Final Box

Fall detection

Some of the examples of fall detection by our method are shown below. It can be seen that our method not only able to localize the person in the thermal image but also reconstructs only the region around the person, which makes it possible to use in the environment having different background artifacts.

alt="fall alt="fall

Dataset

TSF (Thermal Simulated Fall)

It contains 9 videos with normal Activities of Daily Living (ADL) and 35 thermal videos containing falls and other normal activities with a spatial resolution of 640x480.

Please contact Dr. Shehroz S. Khan at [email protected] for access to the dataset. Please specify your affiliation and why you need this data in your email. Download and follow the instructions in the Readme file inside the mrfd folder to perform experiments.

Installation

Conda

Recreate the conda environment using the provided environment.yml file

conda env create -f environment.yml 

Others

These are the major libraries along with their verisons.

  1. keras=2.3.1
  2. tensorflow=1.14.0
  3. opencv=4.2.0
  4. python=3.7.6

We have used tf1.x for the tensorlfow1.x detection zoo model. The provided code consists of person tracking, data processing, model training and testing. The person tracking code must requires tf1.x. If you are using the pre-tracked boxes, you can run the other code with tf2.x as well. Check the Readme file inside the mrfd for more details about pre-tracked boxes.

Additional notes

Citation:

@article{mehta2020motion,
  title={Motion and Region Aware Adversarial Learning for Fall Detection with Thermal Imaging},
  author={Mehta, Vineet and Dhall, Abhinav and Pal, Sujata and Khan, Shehroz},
  journal={arXiv preprint arXiv:2004.08352},
  year={2020}
}

About

This repository contains the source code for the paper Motion and Region Aware Adversarial Learning for Fall Detection with Thermal Imaging

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published