Bring Your Own Data#
This application ships without sample videos or scene data. Before you run install.sh,
prepare the items below.
What You Need#
Input sources: either videos for the 6 camera views, or live RTSP cameras.
A Scenescape scene ZIP (
.zip): your own 3D map with cameras and toll-lane regions already placed on it.The resolution and layout of your cameras, so you can adjust detection regions to match.
1. Add Your Videos#
Place your video files in src/dlstreamer-pipeline-server/videos/ using these exact names:
File name |
Camera view |
|---|---|
|
Front camera, left lane |
|
Front camera, right lane |
|
Rear camera, left lane |
|
Rear camera, right lane |
|
Side profile, lane 1 |
|
Side profile, lane 2 |
Videos must be MPEG-TS (.ts) files. If you use different file names, a different number of
cameras, or live RTSP streams instead of files, update the matching pipeline’s source in
src/dlstreamer-pipeline-server/config.json. See
Perception Layer.
2. Add Your Scene#
The application needs a Scenescape “scene”: a 3D map with cameras and toll-lane regions
calibrated on it, exported as a .zip archive. At startup, the scene-uploader service
imports every <scene>/<scene>.zip found under src/webserver/scenes/.
Build and calibrate your scene using Scenescape’s Create a New Scene and Integrate Cameras and Sensors guides.
Export your scene to
src/webserver/scenes/Smart-Tolling/Smart-Tolling.zipby following Export the Scene. If you keep this exact ZIP name, no further changes are needed. To use a different name, see the step below.
2a. Update ZIP File Name and Corresponding Directory Name (Optional)#
If you kept the ZIP file name Smart-Tolling.zip, then skip this step. If you used a different
name, edit the directory name src/webserver/scenes/Smart-Tolling as well to make sure that
it matches exactly with the name of the ZIP file. Run:
./install.sh smart-tolling
3. Update the Pipeline#
Camera IDs and detection regions in config.json are calibrated for the sample scene. After
swapping in your own videos and scene, update the following for each pipeline in the
Perception Layer:
The camera ID, to match the cameras defined in your new scene.
The detection region (
gvaattachroi), to match your video resolution and camera framing.
4. Models#
When you run init.sh, it downloads the model files listed below from the edge-ai-resources
repository, pinned to a specific fixed commit, and stores them in src/dlstreamer-pipeline-server/models/:
Axle detection
Vehicle type detection
Vehicle front/rear detection
Vehicle color classification
License plate detection
License plate OCR
Models that are already present in that folder are not downloaded again.
If your dataset requires different models, we recommend training them to meet your specific requirements using Geti™ and then integrating them into the application accordingly.