‘AI Guardman’ – A Machine Learning Application that uses Pose Estimation to Detect Shoplifters

Pranav Dar 27 Jun, 2018
2 min read


  • AI Guardman is a machine learning application that detects potential shoplifters
  • The AI is built into security cameras and uses the popular OpenPose technology to estimate the pose of a person and identify suspicious behavior
  • The AI then sends an alert to the shopkeeper’s phone via an application linked to the AI Guardman



The potential of machine learning and AI in the world of surveillance is being tapped into with gusto by companies globally. We have already seen a machine learning project that can spot violence in a crowd of people, which will go a long way towards handling security at events like concerts and sports.

And now a Japanese telecom company has partnered with a startup to build an AI that can be used to detect shoplifters in any store. This AI, being called ‘AI Guardman’, is built into the CCTV cameras that you see in every store these days. Check out the below video that illustrates this technology in action:


How does this technology work, however? The machine learning part of this AI uses a technique called OpenPose. It was developed by researchers at the Carnegie Mellon University to estimate the pose of a person in real-time. It is able to detect a person’s body, hand, and facial points on 2D and 3D images. OpenPose has been met with an overwhelmingly positive response in the ML community and you can download the code to try it out yourself from CMU’s GitHub repository.

There are several pre-defined poses for suspicious behavior. When the AI camera detects such behavior, it instantly send an alert to the shopkeeper via a phone application, also developed for this purpose.

The developers released the results on initial trials and the feedback has been excellent. According to a report in IT Media, they have seen a drop in shoplifting incidents by around 40 percent!


Our take on this

AI is penetrating further and further into the surveillance market. As I mentioned earlier, it’s already being developed for spotting violent behavior in crowds, is being used in China by the police to catch criminals, and even in US homes as a security measure. Facial recognition has also been a game changer in this field – imagine walking past a CCTV camera and being identified in a matter of milliseconds.

It’s a little difficult to image how accurate this particular AI can be. As we saw above, there are pre-defined poses so if someone behaves outside of those parameters, that person might not be labelled as suspicious. Quite an alarming thought, isn’t it?

This does raise challenges of privacy and discrimination. Let me know your thoughts on this technology – are you waiting eagerly for it or are your concerns trumping that excitement? Use the comments section below!


Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!


Pranav Dar 27 Jun, 2018

Senior Editor at Analytics Vidhya. Data visualization practitioner who loves reading and delving deeper into the data science and machine learning arts. Always looking for new ways to improve processes using ML and AI.

Frequently Asked Questions

Lorem ipsum dolor sit amet, consectetur adipiscing elit,

Responses From Readers


Anna 11 Apr, 2022

Interesting post, thanks for sharing.