Camera based intelligent applications are lot of fun! There are many practical applications of it like Industrial Counters, Real Time Object Tracking, Object Classification, Road Traffic Estimation etc. While they are fun and interesting, building them is not that trivial. Generally, building camera based intelligent applications require many modules in the pipelines and a data scientist may not be aware of those. It involves managing hardware, machine learning, a dashboard for user interaction and data visualization and most importantly a way to glue them all together. Taking care of all these requires a considerable amount of time to deploy an application.

I have been working on trying to build tools to simplify and streamline this process and allow data scientists to build such an application in days instead of weeks. In this talk, I will discuss the approach we have taken, the tools that we have built, of which some are open source, and explain how these tools improved the overall time needed to build camera based intelligent applications.

The talk will be structured into 3 parts - 20 mins for the presentation, 10 mins for the hands-on demo to building an application from scratch and the rest of the time for questions.

The presentation is structured as follows with details in slide link provided: 1. Overview 2. Modules in a camera based application a. Data Ingestion b. Pre-processing c. Machine Learning d. Dashboard 3. Our approach a. User agent on data capture device b. A tool to deploy ML functions - Firefly c. Modular dashboard components 4. Demo 5. Questions

In the hands on demo part, I will go through building one image/video based machine learning problem that a data scientist may want to make using our open source modules. I will show how to deploy the application on the rorocloud platform or on your own servers or local machine.

Preliminary slides can be found here