Display Content Capture Alert System
Budget: ₹75,000 – ₹150,000 INR
I want a desktop-based vision solution that watches any TV, computer monitor, or public display in real time and notifies me the moment someone steps in front of it and tries to take a photo. The core may rely on YOLO or straight OpenCV or Any—whichever gives the fastest, most reliable detection.
How it should work
• A camera connected to my desktop continuously analyses the scene.
• When a person holding up a phone (or clearly preparing to photograph) is detected, the system must:
– Crop the frame (or short clip) so the person is centred.
– Timestamp it.
– Dispatch the alert simultaneously to Telegram, WhatsApp, and Email or Any platform.
Key expectations
• Sub-second latency between detection and alert.
• Clean, well-commented source code (Python preferred) plus a requirements.txt and set-up guide so I can recreate the environment.
• Simple JSON or YAML config file to let me adjust confidence thresholds, camera source, and messaging credentials without editing code.
• Tested across the three display types listed above; false positives should be minimised.
• A quick demo video or live screen share proving the system works end-to-end.
Acceptance criteria
1. Running the main script starts live detection with visible bounding boxes.
2. Holding a phone up in front of a TV, monitor, or large public-style display triggers an alert in all three channels within one second.
3. The alert contains the cropped image and readable date/time stamp.
4. Exiting the script cleanly releases the camera and shuts all processes.
If you already have pretrained YOLO weights for “person-with-phone” or can fine-tune quickly, even better; otherwise, propose how you’ll gather or augment data. I’m ready to test as soon as you have an MVP, then iterate together for robustness.
How it should work
• A camera connected to my desktop continuously analyses the scene.
• When a person holding up a phone (or clearly preparing to photograph) is detected, the system must:
– Crop the frame (or short clip) so the person is centred.
– Timestamp it.
– Dispatch the alert simultaneously to Telegram, WhatsApp, and Email or Any platform.
Key expectations
• Sub-second latency between detection and alert.
• Clean, well-commented source code (Python preferred) plus a requirements.txt and set-up guide so I can recreate the environment.
• Simple JSON or YAML config file to let me adjust confidence thresholds, camera source, and messaging credentials without editing code.
• Tested across the three display types listed above; false positives should be minimised.
• A quick demo video or live screen share proving the system works end-to-end.
Acceptance criteria
1. Running the main script starts live detection with visible bounding boxes.
2. Holding a phone up in front of a TV, monitor, or large public-style display triggers an alert in all three channels within one second.
3. The alert contains the cropped image and readable date/time stamp.
4. Exiting the script cleanly releases the camera and shuts all processes.
If you already have pretrained YOLO weights for “person-with-phone” or can fine-tune quickly, even better; otherwise, propose how you’ll gather or augment data. I’m ready to test as soon as you have an MVP, then iterate together for robustness.