AI Detection
On-device AI — runs entirely on your device. No video sent to the cloud. Requires Pro.
SmartRTSP uses on-device machine learning models (TFLite, CoreML, YAMNet). Your camera feeds are never sent to a server for analysis.
Detection Types
Detects people entering the camera's field of view. Configurable detection zone (ROI) to ignore background movement. Uses TFLite object detection model.
Detects cars, trucks, and other vehicles. Useful for driveway monitoring or parking lot surveillance.
Specialized vision model trained to detect flames and smoke in camera feeds. Provides early fire warning.
Audio AI (YAMNet) classifies sounds in real time. Detects: smoke alarm, glass breaking, baby crying. Requires camera microphone.
How It Works — Cascade Detection
Running a full AI model on every video frame would drain battery fast. SmartRTSP uses a two-stage cascade to run AI only when needed.
When a scene is quiet, the heavier model can run less often. Actual CPU, GPU, memory, heat, and battery use vary with the device, model, resolution, frame rate, scene activity, and number of active cameras.
Setting Up Detection
- 1 Open Camera Settings → Detection.
- 2 Enable the detection types you want: Motion / Person / Vehicle / Fire & Smoke / Sound.
- 3 Adjust Sensitivity (Low / Medium / High) to balance accuracy vs false positives.
- 4 Optionally draw a Detection Zone (polygon ROI) to restrict detection to a specific area of the frame.
Do Not Disturb Schedule
Each camera can have a time window during which notifications and recordings are suppressed.
Settings: Camera Settings → Detection → Do Not Disturb. Set start and end times (e.g., 11 PM – 7 AM to suppress overnight alerts while you sleep).
Battery Impact
Measure on the target device and camera mix; there is no reliable universal battery percentage. See Performance & Battery for tuning guidance.