Body posture detection and analysis

Retail industry
Body Posture Detection

Overview

Analysing human body posture can result in many insights like their next moves or possible threats. Especially in super markets we have seen many people shoplifting from difficult angles for cameras. We were tasked to develop a system which based on body postures can flag a person doing suspicious activities.

Technology

  • Deep Learning
  • Computer Vision
  • TensorFlow
  • React JS

Language

  • Python

Key Technical Challenges:

  • Finding body points which can be identified from any angle as part of virtual exo-skeleton.
  • Building deep learning which can be used to do such analytics.
  • Making sure it can detect suspicious actions & flag it.

Business + Technical Points:

  • Analysis & flags must be real time.
  • Should be able to handle 20+ people in a single frame.
  • Must be scalable based on stores

Result:

  • We were able to detect & form exo-skeletons using neural network with real time analytical features.