Public smoking has been an issue for many countries, specially smoking at or near a gas station area. A lit cigarette burns at around 600-degrees Celsius and the gasoline usually has a low auto-ignition point which ultimately means a single cigarette is a switch of a massive explosion.
Even though people knew these facts, they still smoke while pumping gas. Around 50 people get hurt as the result of a gas station fire each year. To prevent this problems, Microsoft developed an AI-powered alarm system that detect and stop smokers before a catastrophe.
Microsoft will installed cameras on gas stations which observe people and whenever certain behaviors like lighting a cigarette arose, the onboard AI flags the image and uploads it to the cloud where more robust AI determines if an alarm should be sounded.
For the implementation of this AI-powered alarm system Microsoft has made partnership with Oil company Royal Dutch Shell. Microsoft published a blog post detailing a pilot project at two Shell gas stations, one in Thailand and one in Singapore.
Microsoft didn’t talk much about the alarm but hopefully it’s something ridiculing like a profanity-laden rant from Samuel L. Jackson. I’m not usually for shaming people, but just about everyone needs to get gasoline, we shouldn’t all have to risk our lives because someone’s addiction can’t wait five minutes.
The key to the quick turnaround is edge computing, which processes data near its site of origin rather than in the cloud. In this example, which uses Microsoft’s Azure IoT Edge platform, the footage analysis happens onsite at the gas stations. Only frames that raise any red flags move on to the cloud for advanced processing.
“Each of our retail locations has maybe six cameras and captures something in the region of 200 megabytes per second of data,” Daniel Jeavons, Shell’s general manager for data science, told Microsoft. “If you try to load all that into the cloud, that quickly becomes vastly unmanageable at scale. The intelligent edge allows us to be selective about the data we pass up to the cloud.”
Microsoft’s solution is to use simpler algorithms to detect possible smokers at the pump, and then send those frames to the cloud for analysis. To make it work perfectly, Microsoft developed a low-power image recognition system that could run locally within a framework that was capable of working with the cloud.
The program is still in its initial stages. Microsoft promises that they will soon finished there final operations and if there two experimental project one in Singapore and the other in Thailand get successful, the company will add it to everywhere. We hope to see similar tech for applications far beyond the pump, according to the post, like predicting equipment failures in industrial settings.
You can read more about Microsoft’s partnership with Shell on the company’s AI blog.
More in AI :