The problem JhumriCam solves
We wanted a fun yet functional way to make sense of whoβs in the office people, cats, and chaos included.
JhumriCam taps into our office RTSP cameras to:
- π₯ Count how many people are physically in
- π Detect when our office cat Jhumri makes an appearance
- π₯ Track who visits the office the most (yes, we made a foodie leaderboard)
- π€ Compare predicted lunch attendance (based on π counts) vs actual headcount from the cameras
- π¬ Share all of this via a Telegram bot with /headcount, /start, etc.
Challenges we ran into
-
Getting access to RTSP in office.
-
Model detection
-
RTSP setup with multiple sources: We initially struggled to simulate and test with live camera feeds. As a workaround, we created RTSP streams using MediaMTX and local videos to simulate cameras until real feeds were available.
-
Low voltage issues on Raspberry Pi: While experimenting with physical camera setups, we encountered hardware limitations and power drops. Disconnecting peripherals helped stabilize it.
-
Python OpenCV + RTSP instability: Handling multiple streams simultaneously required retries, fallbacks, and frame decoding tweaks.
-
Telegram bot in group chats: Telegram bots donβt listen to non-command messages in groups, so we implemented slash commands like /headcount and /errorrate to enable functionality everywhere.
Despite the hurdles, we now have a quirky, lovable surveillance companion β powered by Jhumri πΎ
Technologies used
Cheer Project
Cheering for a project means supporting a project you like with as little as 0.0025 ETH. Right now, you can Cheer using ETH on Arbitrum, Optimism and Base.