Realtime Video Analytics using Python - From Edge to Cloud and Beyond



Real-time video analytics requires high computation power as it crunches multiple frames per second from a video stream. The task becomes complex when it has to run on a low compute edge device and still ensure that the processing rate catches up to the camera FPS. Added to this deployment, management and tracking of edge solution need to be planned as well. Here, we will talk about our experience in building a product on python - a realtime video analytics pipeline (for ANPR) that runs on an edge device.

This talk will cover the following aspects:

  • End-to-End video analytics pipeline
  • ML Model, PyCUDA and TensorRT
  • IoT - Provisioning, Security, OTA Updates and Monitoring

Here is the breakup of talk with appr. timings:

  • Introduction - 3 mins
  • Pipeline overview - 6 mins
  • ML Model - 3 mins
  • PyCUDA and TensorRT - 6 mins
  • IoT: Provisioning & Security - 4 mins
  • IoT: OTA updates & Monitoring - 3 mins
  • Q & A - 5 mins


  • Basic knowledge of python
  • Basic knowledge of cloud services

Video URL:

Content URLs:

Initial draft of presentation is available here -

Speaker Info:

Hi, I'm Vivek. I have nearly 6 years of experience in python. Started my career as a backend engineer and moved to ML domain. I have worked across automotive, energy, billing & payments and FinTech verticals. I currently work as an ML Engineer in Toyota Connected India. I'm passionate to work with autonomous vehicles, embedded devices, hardware platforms, CV and linux. Keen to explore new things and learn from failures.

Speaker Links:


Section: Embedded Python and IOT
Type: Talks
Target Audience: Intermediate
Last Updated: