Remix.run Logo
maccard 3 hours ago

What LLM can I get and feed hundreds of hours of video into that will give me the position of a specific vehicle alongside when that happened?

An LLM isn’t going to help you here, but basic Computer Vision and a SQL database has been a solution _if you have the cameras_. I wrote a license plate reader as a university project using OpenCV almost 20 years ago.

handoflixue 2 hours ago | parent | next [-]

If you were able to write one 20 years ago, I dare say an LLM could whip one up super-fast. Or just search the internet and tell you where to find one.

One of the risks of LLMs is that a lot of tasks go from "an expert could do this easily given a few weeks" to "anyone who thinks to ask an LLM can do this easily and get results the same day"

jubilanti 2 hours ago | parent [-]

[flagged]

Manuel_D an hour ago | parent | next [-]

Automated License plate readers are a half century old at this point: https://en.wikipedia.org/wiki/Automatic_number-plate_recogni...

tough 2 hours ago | parent | prev | next [-]

why wouldn't codex or claude just reach for whatever FOSS https://github.com/openalpr/openalpr

jubilanti 2 hours ago | parent [-]

[flagged]

microgpt 2 hours ago | parent | next [-]

Yes, an ALPR is basically just a glorified [thing that an ALPR does]

tough an hour ago | parent | prev | next [-]

Dropbox stock is trading at 50% of its initial price 5y ago when it went public, maybe the public markets also don't understand the difference between rsync and Dropbox.

SR2Z 2 hours ago | parent | prev | next [-]

I honestly can't tell if this is ragebait or you believe this.

My friend, if you have a database of license plates extracted from single images taken by multiple cameras, YOU ARE TRACKING UNIQUE VEHICLES ACROSS A REGION.

Terabytes of data don't matter because you don't need to search terabytes, you need to search a few MB of text data. You don't even have to store the original video.

assimpleaspossi an hour ago | parent [-]

But you don't have such a database.

therealdrag0 2 hours ago | parent | prev | next [-]

You’re moving the goalposts. The original point of this thread is that Flock AI technology is hardly needed to efficiently search traditional video footage for license plates.

malcolmgreaves an hour ago | parent | prev [-]

Ok, so you don’t understand that ALPR is two commodity technologies: object detection and OCR.

nullsanity 2 hours ago | parent | prev | next [-]

[dead]

handoflixue 2 hours ago | parent | prev [-]

> Assuming they have access to terabytes of regional video surveillance but don't have their own compute besides what the LLM will buy for them

Amazing how you can move the goalposts to make things impossible, isn't it? Where in the world did "without compute" come from? Are they not even allowed a decent desktop computer?

IncreasePosts an hour ago | parent | prev [-]

Any competent llm would write a script using opencv to extract the license plates.

I did this with Gemini 3, mostly for fun and to test it's capabilities. Teslausb records all dash cam videos and auto syncs it to my nas when in wifi range. Yolo and opencv extracts and does ocr on any defected license plate, and puts it all on a map, along with trip information. Not particularly useful or interesting, and not something I would have done pre-llms, but the difficulty was basically writing a one paragraph prompt and using some free tokens