π WildDetector Guide
WildDetector is an Android app that uses on-device AI to detect wildlife β by sound and by sight β in real time. It works completely offline: no data is sent to any server during detection. When you are signed in, your detections can optionally be saved to the cloud and shown on a shared map.
πΎ Species
When you open the app you choose which species to monitor. Each species has its own AI models and tuned settings.
Pollinators (binary)
Audio CNN + Motion detectorDetects pollinator presence by wing-beat frequency (~200β300 Hz). Binary: present or not. Works best 0.5β2 m from the hive or flowers.
Amphibians (multiclass)
Audio CNNIdentifies calls from multiple amphibian species. Multiclass: identifies the specific species detected. Works best at night near water sources.
Avian (binary)
Audio CNN + AI visualDetects bird presence by song and visual. Binary: present or not. Can detect presence even without a visual.
Reptiles (binary)
AI visualVisual detection of reptiles at 10β15 FPS. Binary: present or not. Point the camera at the subject.
General
Anomaly + Motion onlyNo species-specific model. Uses anomaly detection and motion tracking to catch any unusual activity. Ideal for unknown species or custom field setups.
π¬ How Detection Works
WildDetector fuses three independent detection sources. Each can trigger independently or be chained in a pipeline.
Audio Classification
The microphone continuously captures audio. Every ~1 second, a CNN model classifies the sound and produces a probability score. When the score exceeds the threshold for several consecutive windows, a detection is fired.
AI Visual Detection
The camera runs an on-device AI visual model on every frame. When the target species appears in multiple consecutive frames (configurable), a detection is triggered. Bounding boxes are drawn live on screen.
Anomaly Detection
An autoencoder model learns what 'normal' sounds and images look like during a calibration phase (~2 minutes). Once calibrated, anything outside the normal distribution triggers an anomaly alert β useful for catching rare events.
Motion Detection
A lightweight frame-differencing algorithm detects moving objects in the camera view. Motion is a trigger-only source β it wakes up other detectors but does not generate a standalone detection by itself.
π Detection Pipeline
By default every source fires independently. The pipeline lets you chain sources so that a detection is only recorded when multiple sources confirm the event in sequence. This dramatically reduces false positives.
# Example: require motion then audio
Stage 1 β Motion detected
Stage 2 β Audio confirms within 10 s
β Detection recorded β
You can configure the pipeline in Settings β Pipeline Editor. Each stage has a 10-second timeout β if the next stage does not fire in time, the pipeline resets.
βοΈ Settings
ποΈ Audio
Turns the audio CNN classifier on or off. Disable if you only need visual detection to save battery.
Select the microphone to use. Useful if you connect an external USB microphone for better range.
Left, Right, or Mixed. Use Left or Right when using a stereo directional microphone.
How many consecutive high-probability windows are required before firing. Higher = fewer false positives. Default: 3.
π· Camera / YOLO
Turns on the on-device AI visual model. Uses more CPU/GPU and battery.
Minimum confidence (0β1) for the AI visual model to consider a detection valid. Default: 0.4.
How many consecutive frames must contain the target before firing. Default: 3.
Saves an annotated JPEG snapshot with boxes drawn on it to your gallery each time the AI visual detector fires. Premium only.
β‘ Anomaly Detection
Trains an autoencoder on ambient sound during the first ~2 minutes, then alerts on anything unusual.
Same as audio anomaly but for the camera feed. Useful in fixed-camera setups.
Automatically adapts based on your feedback (thumbs up / thumbs down on each anomaly card).
π‘ Motion
How easily motion is detected. High sensitivity picks up insects; low sensitivity ignores leaves and camera shake.
Maximum size of a detected moving object. Set small to detect insects; large to detect mammals.
Minimum speed a blob must travel to count as motion. Filters out slow camera drift.
Draws the path of each moving object on screen for visual feedback.
π¬ Video Recording
Automatically records a video clip around each detection event. Premium only.
How many seconds before the detection to include in the clip. Default: 5 s.
How many seconds after the detection to continue recording. Default: 5 s.
π Battery & Screen
Prevents the screen from turning off during a detection session. Recommended for field use.
On Samsung devices, add WildDetector to Settings β Battery β Never sleeping apps to ensure background detection keeps running with the screen off.
πΎ Presets
Saves all current settings under a name so you can switch between configurations quickly.
Restores a previously saved configuration.
π€ Account & Premium
WildDetector has three access levels. You can start immediately as a Guest with no account required.
Guest
- β No sign-in required
- β General detector only
- β 3-minute detection sessions
- β GPS must be enabled
- β Local session log
- β Nothing saved to cloud
Free Account
- β Sign in with Google
- β All species detectors
- β 3-minute detection sessions
- β GPS location logged
- β Detections on live public map
- β My Detections history
Premium
- β All species detectors
- β Unlimited detection sessions
- β Save video clips & photos
- β GPS location logged
- β Detections on live public map
- β My Detections history
- β $10 / month or $80 / year
Sign in with Google from the species selector screen (tap the account button at the bottom). Subscriptions are managed through Google Play and can be cancelled at any time. Premium status is cached for up to 35 days offline (monthly) or 370 days (annual).
π My Detections
Every detection made while signed in is saved to your personal history, organized by session.
πΊοΈ Live Dashboard
The WildDetector dashboard shows real-time wildlife detections captured by the app. On the map you can see where bees, frogs, birds, reptiles, and other species have been reported. You also have a feed view with the detection history, and a devices view showing which phones are currently active.
Open the dashboard at wild-dashboard.vercel.app
π Community Presets
Presets let you save your current detection settings under a name and share them with other users. Free users can browse and apply public presets from the community.
π‘ Tips & Tricks
β Use an external USB microphone
A directional or parabolic microphone connected via USB OTG dramatically increases detection range. Select it in Settings β Audio device.
β Set pipeline for fewer false positives
Chain Motion β Audio or Motion β AI visual so the app only records confirmed events, not background noise.
β Calibrate anomaly detection first
Start a session and wait 2 minutes without the target species present. The anomaly detector learns the background and becomes much more accurate.
β Use pre-roll to never miss the moment
Set a 10-second pre-roll so that when a detection fires, the video clip includes the approach, not just the aftermath.
β Samsung users: add to Never Sleeping Apps
Without this, Samsung's aggressive battery manager kills the background service within minutes of the screen turning off.
β GPS must be enabled
The app requires location access to tag detections with coordinates. Make sure GPS is on before starting a session.