DDOM (Data Driven Online Monitoring) System

Description

This repo is a ROS Package which can be used for monitoring camera based Human Detection Modules.

Functioning

1) The Framework consists of three components

• Position Deviation Estimation (PDE) :

  • Takes input of bounding boxes from the human detection model.
  • Creates a tracklet of the nearest person from the camera.
  • Estimates Deviation in tracklet of the nearest person detection by the human detection model.
  • Outputs the Estimated Deviation into a ROS topic (as a float message).

• False Negative Monitor (FNM)

  • Takes RGB image input from a camera.
  • Classifies negative instances (no detection instances) from the human detection model into False Negative and True Negative.
  • Outputs the binary classification into a ROS topic (as a bool message).

• Detection Continuity Checker (DCC) :

  • Checks for continuity of output from the detection model.
  • Creates an alert notifying the non-continuity in output.
  • Changes the Active Monitoring Component between FNM and PDE.

Dependency

• The Noetic branch of the repo depends on :

  • SENTOR System: https://github.com/lcas/sentor
  • SORT Tracker: https://github.com/abewley/sort

• The SENTOR is used as the DCC for monitoring the human detection ros topic and publishes a boolean ROS message based on the continuity of detections from the human detection mdoel

• The SORT is used as the tracker in the framework for generating the tacklet of the nearest person infront of the camera.

INSTALLATION AND USAGE

• To install the framework :
  • clone the repo into a ROS workspace

  cd /path/to/workspace/src
  git clone https://github.com/hariharan20/ddom -b noetic 
  

  • install the dependencies :

  cd /path/to/workspace/src
  git clone https://github.com/lcas/sentor
  cd DDOM/tracker/src
  git clone https://github.com/abewley/sort
  cd /path/to/workspace/
  

  • build the workspace

  catkin_make
  

  • source the workspace

  source /path/to/workspace/devel/setup.bash
  

• Launch the framework :

roslaunch ddom ddom_v2.launch 

• Topic names and path to models are set in a config file located at
  /path/to/workspace/src/DDOm/ddom/config/config_v2.yaml
• Change the config file for custom topic names and camera-specific parameters.

If you find this work useful, please refer to the following citation:

@inproceedings{arunachalam2024runtime,
  title={Runtime Anomaly Monitoring of Human Perception Models for Robotic Systems},
  author={Arunachalam, H and Huang, Z and Hanheide, M and Guevara, L},
  booktitle={2024 IEEE 20th International Conference on Automation Science and Engineering (CASE)},
  pages={723--729},
  year={2024},
  organization={IEEE}
}