AI Motion-Alert CCTV Software
Budget: $750 – $1,500 USD
I need a stand-alone CCTV monitoring application that focuses on smart motion detection and instant alerting rather than basic live-viewing or simple recording. The core requirement is an algorithm that can reliably recognise and differentiate humans, vehicles, and wildlife in real time and then trigger configurable alerts (push alerts with a picture of the detected to a local pc either via web based or program based)
Key points you should know
We have 768 cameras currently to add to the system
The system should be hub based and the hubs should do the analytics and send the push notification photos to the main pc
Should be able to load 40 camera per hub depending on the specs of the hub pc
• Video input will arrive over standard RTSP / ONVIF streams from existing IP cameras.
• Accuracy and speed matter more than fancy UI elements; a lean desktop or web dashboard that shows the current feed, bounding boxes, and an alert log is enough.
• OpenCV, TensorFlow or a comparable framework is fine; I am open to the model you recommend as long as false positives remain low in typical indoor-outdoor lighting.
• Platform can be Windows, Linux, or a cross-platform container—choose what lets you ship fastest without licensing headaches.
Selected detection modes (mandatory)
• Human detection
• Vehicle detection
• Wildlife detection
Deliverables
1. Compiled application or Docker image ready to run on my server.
2. Source code with a short README covering install steps and config options.
3. JSON or REST endpoint that pushes alert data (timestamp, object class, confidence, snapshot path).
4. A small test video set plus metrics showing detection accuracy on each class.
Acceptance criteria
• Under 2-second end-to-end latency from motion to alert on a 1080p stream.
• <5 % false-positive rate in daylight; <10 % at night when IR is active.
• Clean hand-off: after I change camera RTSP URLs in a config file, the system must reconnect without code changes.
Once you confirm the overall approach, I will supply a couple of sample streams so you can fine-tune.
Key points you should know
We have 768 cameras currently to add to the system
The system should be hub based and the hubs should do the analytics and send the push notification photos to the main pc
Should be able to load 40 camera per hub depending on the specs of the hub pc
• Video input will arrive over standard RTSP / ONVIF streams from existing IP cameras.
• Accuracy and speed matter more than fancy UI elements; a lean desktop or web dashboard that shows the current feed, bounding boxes, and an alert log is enough.
• OpenCV, TensorFlow or a comparable framework is fine; I am open to the model you recommend as long as false positives remain low in typical indoor-outdoor lighting.
• Platform can be Windows, Linux, or a cross-platform container—choose what lets you ship fastest without licensing headaches.
Selected detection modes (mandatory)
• Human detection
• Vehicle detection
• Wildlife detection
Deliverables
1. Compiled application or Docker image ready to run on my server.
2. Source code with a short README covering install steps and config options.
3. JSON or REST endpoint that pushes alert data (timestamp, object class, confidence, snapshot path).
4. A small test video set plus metrics showing detection accuracy on each class.
Acceptance criteria
• Under 2-second end-to-end latency from motion to alert on a 1080p stream.
• <5 % false-positive rate in daylight; <10 % at night when IR is active.
• Clean hand-off: after I change camera RTSP URLs in a config file, the system must reconnect without code changes.
Once you confirm the overall approach, I will supply a couple of sample streams so you can fine-tune.