The primary goal of this project is to analyze a live video stream from the main viewing window of the Ocean Voyager tank at the Georgia aquarium, and apply real-time object detection to locate and identify the visible animals. It also includes some additional related functionalities, such as the ability to analyze a video file (rather than a livestream), the ability to periodically capture a frame and save it as a jpg without attempting to analyze it (useful for collecting images for annotation), and various visualizations (both live visualizations for demo purposes, and more complex visualizations that are instead saved to disk).
Included is a yolov11m model (yolov11m_ov.pt) trained to detect the following species: [whale shark, bowmouth guitarfish, guest, green sea turtle, diver, giant grouper, zebra shark, tarpon, white-spotted guitarfish, spotted eagle ray, sandbar shark, blacktip reef shark]
ovlive_sample.mp4
Note that to run the demo you will likely need to update the hard-coded "stream_url" variable, as it changes periodically. See the below image for how to find the stream url in the network section of the browser inspection panel / DevTools panel of Google Chrome. You want the Request URL for the most recent "chunklist*.m3u8" request.
