Release Date: 09.11.2022
With the help of measurements in our Performance Lab, the accuracy improved significantly: 95% was achieved for the main classes (car, truck, bus); 85% for the subclasses (van, truck, semitrailer).
On top, the description and distinguishment of the single unit truck and articulated truck have been adapted based on feedback and input from existing installations.
Furthermore, we introduce the class "others" with this Release. Any vehicles in our training dataset which can not be allocated to any of the classes/subclasses, according to their definition, are classified as others. This class consists mostly of agricultural tractors and special big construction vehicles.
With the Multi-Role support in the SWARM Control Center, three different user access rights are available. This allows you to customize the rights per user to follow your existing policies, e.g. some users can only view the information and have no editing rights.
Three different user types have been added. The admin can edit everything and create Monitoring alerts, the standard user is allowed to edit any configuration and dashboard and the viewers are allowed to view all the dashboards in Data Analytics.
Additional support for configuring your use cases in the best way. With our new feature Track calibration, it will be much easier to decide where to place the event types, for example, your Counting line. A sufficient amount of past tracks will be combined to an overlay in the calibration frame. This makes it clear to you, where the objects are detected the best in the camera frame. So the CL can be placed accordingly, and it will be easier to get the best output result from the first day on.
E2E monitoring services have been added! Some of them are available in the background so that we can provide high reliability for our SWARM Perception Boxes and Control Center.
Speaking of Control Center: There is a new feature to receive email notifications regarding issues with the Perception Boxes. These monitoring alerts can be set individually by administrators.
Find more information in the following section of the documentation:
Traffic & Parking model improvements
In this Release, we were able to further improve our detection model stability for traffic and parking use cases, especially in bad weather situations with heavy rain.
Two concrete examples of areas in the model have been improved:
- On Parking Entry-Exit use cases the model is improved especially on top-down scenes in rainy conditions and can now recognize these difficult cases stable. In our Performance lab scenarios, the Parking Entry-Exit accuracy is now at 99% taking into consideration the installation specification that is followed.
- For vulnerable road users improvements focused on people pushing a bicycle who are considered as people in the Swarm class definition have been implemented.
ANPR accuracy improvement
With a performance improvement on the processing side, an accuracy improvement for our ANPR case has been achieved. In our performance laboratory, we have improved the performance from 90% to ~93%
Update size reduction
The download size of the updates has been reduced by a factor of 5 and will be around 100MB for the update. This will reduce the data traffic especially relevant to installation with LTE connectivity.
Data Analytics improvements
- Timezone handling: In a given dashboard any timeframe is now displayed in the configured timezone. This includes as well the API calls and csv exports.
- Dashboard reloads: The reload of all widgets is triggered with any refresh of the page as well if you exit a dashboard and enter it again.