Building Reliable Vision AI Solutions | Eric Clark, SpringML

Google Cloud – Partner profile 

Topics: Vision AI, Video, Edge Solution, Machine Learning, Google

Transcript Excerpt

This video demonstrates what it takes to create a reliable vision AI. Most people do not consider videos to be a data source. They are capable of far more than just surveillance and safety-related applications. Data is abundant in videos, and cameras are ubiquitous. Cameras are in our pockets, cell phones, cars, and buses.  In this instance, let us consider transportation. Police cameras are placed at major intersections to monitor traffic conditions. Instead of just monitoring those conditions with cameras, live AI looks for changes in traffic. If an incident occurs at an intersection that was not being monitored by humans. 

Vision AI can intervene and respond much more quickly in that situation. It’s just one of many examples of what this technology can do. The problem right now is that Google has already built several vision AI models using API. The task is to figure out where to get the videos, and we have three options. One is video/archive storage, another is Vision AI on Streaming Video, and the third is Edge Solutions.  

Edge Solutions are situations in which we can have a variety of cameras in fixed or mobile locations and must use edge or hybrid processing to collect them before sending them to the cloud for processing. Videos are large files. Because of their resources, it is not the easiest to deal with. When it comes to operationalizing those models, storage and processing present some challenges.  

According to the Machine Learning Methodology, one of the most difficult aspects is collecting data in a reliable and automatic manner. Another reason vision AI from edge sources is more difficult is that it requires a combination of hardware, software, and connectivity.  

To address this issue, Eric and his team created an end-to-end Vision AI technology using Google technology that assists customers in deploying production models. On the hardware side, they used Google coral technology and developed software that runs on this hardware using vision API, video intelligence, and finally insights and actions where they use Big Query, App Engine, and Google Maps to review and respond to results.

Enable Group – Microsoft Research 

Enabling Physical Analytics in Retail Stores Using Smart Glasses