-
Notifications
You must be signed in to change notification settings - Fork 1
Configuration Variables
Winfried Holzapfel edited this page Oct 27, 2019
·
1 revision
Each device has a number of configuration variables that can be adapted to particular scenarios. They are sent to the device as a response to the next update request sent from the device. These variables can be configured from the configure menu available under the device listing on the dashboard.
| Name | Default Value | Purpose |
|---|---|---|
| uuid | Device Dependent | How the device is identified with the server. Never change to get reliable connection |
| frequency | 20 | How often the device sends updates over the network |
| image_quality | 25 | The JPEG compression quality. 100 indicates lossless transmission. |
| model | dnn/yolov3.weights | Path to the file including weights for the neural network (darknet). |
| config | dnn/config.cfg | Path to the configuration file describing the neural network. |
| confidence | 0.2 | Percentage of how sure we need to be to flag a human. Increase this if there are false detections. |
| nmsThreshold | 0.1 | How close bounding boxes need to be before they are considered the same person, expressed in normalized total box area |
| skipFrames | 5 | Number of frames to skip between detections when using the tracking subsystem |
| imageWidth | 320 | Width of the captured image. Influences tracking performance. |
| imageHeight | 240 | Height of the captured image. Influences tracking performance. |
| useOpenCL | 1 | Whether to use OpenCL acceleration for neural net computation on supported platforms |
| useCSRT | 0 | Whether to use KCF (default) or the more accurate but slower CSRT tracker |
| neuralNetQuality | 416 | Pixel dimensions of the neural net. No need to change under normal circumstances |
| maxDisappeared | 50 | Time in frames before a person is considered 'lost' from its marker |
| searchDistance | 50 | Area to search for a person around the original marker for subsequent detections |
| useTracking | 1 | Whether to use the slow but nice-looking tracking subroutine. Not available on IoT platforms |