In an emergency every second counts. First responders work extremely hard to ensure they can be on the scene of an accident in a matter of minutes. Only after an incident is reported can first responders head to the scene. This relies on the people involved being in a well enough state to make the call, or a bystander to make the call.

What if we can eliminate the time of informing first responders of an incident?

We made an AI that can detect car crashes at intersections, informing first responders of a crash faster than people can call it in.

We ingest a video (from a live stream later) and analyze one frame per second of said video, allowing for multiple streams of videos on the same server.

The photos are initially ingested by a stream analyzer which detects vehicles in a frame. If it detects two or more vehicles overlapping, it sends it to the next step, if not it gets discarded.

When an overlap is found the photo is sent through our second neural net which was trained on thousands of images of car crashes extracted from videos. At this step we would inform the dispatcher of a possible collision and they can forward the information to first responders.

We are using AI and security public security cameras to detect and inform first responders of vehicle accidents decreasing first responders’ response time and increasing our city’s safety.

Built With

Share this project: