Interactive camera application for selective object detection. Real-time object counting is available in user-defined areas of the video image. The application can differentiate between vehicles such as trucks, cars, buses, and motorcycles.
Users can interact with the live video and select areas of interest. Object detection and counting will be isolated to the detection zones specified by the user. Multiple areas can be defined, and the number of objects will be displayed for each zone.
The Revis box is an End-to-End solution that can be configured for your needs. This means that even though parts can be exchanged, we provide a turnkey solution. While we rely on NVIDIA devices, the camera itself can be selected depending on the use case. This has several advantages such as extending existing cameras with AI functionality or selecting a special camera, for instance, night vision.
To use the app on your own devices:
- Register an NVIDIA Jetson PC in your device swarm on this platform
- Plug any USB-Cameras into your device
- Access your device in the swarm and install this app on it
To create a multi camera setup, just repeat the steps above with as many devices as you like.
The detection data of all cameras in your swarm will be available in the app dashboard.
This app provides a dashboard summarizing detection data from all swarm cameras into a common view. Collected data histories, such as the number of detected objects per class and the current video frame, will be displayed in the dashboard. The history of detections is persisted, providing a summary in different graphs.
The detection data history is collected in the cloud storage backend of the app. Each swarm automatically uses its own private cloud data storage. This ensures that the app dashboard in a swarm can only access and show the data of that swarm. This is how client data separation is reliably and securely enforced.
The video snapshot that is shown in the dashboard is recorded historically with only one frame per second while the live stream (see below) has a higher frame rate and lower latency.
The app additionally provides a web interface on each individual camera that allows users to view the low latency live video stream from that camera. This interface also allows users to configure the video and detection model for the camera.
Each camera can also be configured by external parameters in the app settings on the device. These settings can be given on device or group level for the whole swarm at the same time.
| Parameter | Default | Description |
|---|---|---|
| Model | Yolov8s | Select a version for one of the YOLO Object Detection Models |
| Detection Classes | 0,2,3,5,7 | A list of class ids from the COCO dataset. If left empty, then all classes are used. See here |
| Confidence | 0.1 | (0 - 1) Sets the minimum confidence threshold for detections. Objects detected with confidence below this threshold will be disregarded. Adjusting this value can help reduce false positives. |
| Intersection over Union | 0.7 | (0 - 1) Intersection Over Union (IoU) threshold for Non-Maximum Suppression (NMS). Lower values result in fewer detections by eliminating overlapping boxes, useful for reducing duplicates. |
| Width | 1280 | Horizontal Resolution in pixel of the camera stream. The camera should support the selected resolution. |
| Height | 720 | Vertical Resolution in pixel of the camera stream. The camera should support the selected resolution. |
This app can currently only be used NVIDIA Jetson systems.
Any USB or IP camera connected to the Nvidia PC can be used with the app.