FAQs
Frequently Asked Questions
In this section you will find all frequently asked questions. We hope to be able to support you with this. This area will be updated continuously.
1. Solutions
What are the requirements for using your product?
Camera(s): This is essential, of course. The number of IP cameras needed entirely depends on the use case. You can find more details regarding camera specifications in our technical documentation. We are happy to support you with this and cameras can be delivered directly via Swarm Analytics.
Electricity: Naturally, a power supply is needed. This can be provided by either a barrel jack power adapter (i.e. for our SWARM Perception Box P101), a 230V power connection (speaking of the SWARM Outdoor Perception Box OP101AC), or even via solar- or battery-powered systems, for which our OP101DC is particularly well suited.
Internet: An internet connection is mandatory to get the system up and running. Without an internet connection, one cannot access the system and any configuration will be impossible. The connection can be done by either cable (LAN) or mobile connection (LTE).
2. Product: Sensor
How many camera streams can be handled with one box?
For a detailed answer to this question, please visit the related section in our technical documentation here.
Which objects can be recognized? How does the detection work exactly?
Objects such as vehicles, cyclists or people are detected and classified by models pre-trained by us. On the technical side, the SWARM Computer Vision software first looks at individual frames of the video stream in real time. The image is analyzed by an artificial neural network and relevant objects (vehicles, pedestrians, etc.) are detected.
In a second step, the objects are classified more precisely, e.g. to distinguish cars from trucks or motorcycles from bicycles. Subsequently, the objects are combined over several frames to detect the movement of objects. These so-called ‘tracks’ are used to perform relevant events and counts with the SWARM Event Engine. This data is generated by so-called event triggers and encrypted. Only the anonymized data without inference to the original camera image leaves the device and can be used for evaluation and analysis. The main advantages of this Computer Vision approach are its flexibility and extensibility.
All information on the camera image that is visible to humans could thus also be used for automated analysis. For example, in order to detect new objects, no additional sensor is required, but only a software update - in terms of scalability and adaptation to possible further requirements, this is a major advantage over other approaches.
How and where are the parameters processed?
The generated video from the camera is processed exclusively on the Perception Boxes. No video data is transferred to a server/cloud or stored. The Perception Box is connected to a suitable camera on site. Events from the configured event types are transmitted to the Azure Cloud via MQTT and stored in a database there. This enables visualization and evaluation centrally and conveniently in the browser of your choice via SWARM Data Analytics for all Perception Boxes.
Alternatively, events can be transmitted via an MQTT server provided by the customer. In this case, further processing of the raw event data is the responsibility of the customer and enables even more use-cases and custom integrations.
Which vehicle classes are recognized and how are they defined?
With the SWARM software, multiple motorized as well as non-motorized traffic classes can be surveyed - from cars to trucks with trailers to pedestrians. For motorized traffic in Germany, we have followed the BAST classification guidelines as far as visually possible, but we also offer other standards. More details regarding the object classes can be found in our technical documentation.
How about my other custom class: Can you train on that?
We understand that custom classes can play a crucial role in meeting specific use-case requirements. Our dedication to delivering personalized solutions for our customers and partners is paramount, and we constantly strive to enhance our product and cater to the distinctive needs. Nevertheless, it’s important to acknowledge that the progress of mobility and traffic behaviors and developments significantly influence the types of classes we teach our models to recognize. Therefore, we keep a close eye on emerging mobility trends and adapt our solution accordingly, most recently by adding e-scooters as an additional class.
We appreciate your understanding that training new classes is a complex process that requires extensive research and testing to ensure that our models perform accurately and reliably with our known level of quality. If you have a relevant use-case in mind, feel free to get in touch with our Sales team to discuss the opportunities and scaling of the project.
How accurate are you?
Accuracy varies for each use case and is, of course, also dependent on environmental conditions, no matter how precisely the Swarm algorithm works.
Nevertheless, we can work with an accuracy between 95 and 99% with one of the highest accuracies in the industry for standardized applications, such as parking lots or traffic counting on designated highways and urban streets.
See the next questions for further information and support. For further information regarding our technology’s accuracy, we recommend this section in our technical documentation.
How can I improve my accuracy? Which environmental conditions have to be met in order to get great results?
Our technical documentation provides help for configuring the respective use case:
To what extent is detection limited at night and/or in bad weather (i.e. heavy rain, fog, etc.)?
At night, the accuracy of detection depends on the available lighting and camera settings. With sufficient lighting and camera settings according to our specifications, the loss of accuracy can be reduced to a minimum. Heavy rain, fog as well as other extreme weather situations can be partly compensated.
Basically, for all of the above scenarios, anything that can be clearly identified by the human eye can also be detected by the software. If you have example videos of your use-cases, we can test them through our software and measure the accuracy over a predefined period of time.
Do you have any recommendations for cameras?
The following sections in our technical documentation provide configuration details and recommendations for cameras depending on the respective use case:
3. Product: Control Center
3.1. Device Configuration
Is it important that the entire vehicle or object travels “inside” the zone or is it sufficient if the center-point of the vehicle does so?
The center-point of the vehicle (or other object) is tracked, therefore it is important that the center-point of the vehicle moves in the entry zone and leaves in the exit zone. Bear in mind that the camera view is a 2D representation of the 3D world, so the zones often need to be larger than you expect. The center-point of the vehicle may be above road level.
Is it okay if the vehicles stand in a zone for a while, i.e. can I create a zone “in front” of the traffic light so that the vehicles are still in the “entry” zone?
It is okay if vehicles (or other objects) stand still for a while in a zone. When they start moving again, they will continue to be tracked. If possible, it is best to avoid zones that cover parked vehicles. This can cause performance problems as vehicles are constantly tracked and can be confused with each other if another similar vehicle passes nearby.
Can zones overlap or be very close to each other?
Yes, zones can be very close to each other or overlap. However, an object (e.g. a car) must be detected as entering and exiting two different zones (i.e., without overlapping) for its track to be recorded.
Is it only possible to gather entry and exit data, or also the density of groups of people?
Entry and exit counts are very useful for identifying the total number of people in a specific area (e.g. outdoor pool, market, or stadium). If COVID-related spacing rules are to be followed, the distribution of people is important. Therefore, the Swarm Analytics device also directly outputs the distribution of objects in an area. However, partial roofing or large sunshades can make it difficult to detect the distribution of people.
4. Product: Support and Maintenance
What are the next steps after my order?
As soon as you receive your order, you can start with the installation process. For example, the installation and configuration of the SWARM Outdoor Perception Box is simple and can be completed in 30 to 60 minutes. The following requirements are necessary for an outdoor installation, additional to the box itself:
Suitable IP camera with PoE LAN cable
Continuous 230V power supply
Standard miniSIM card with more than 600MB data volume per month
Screwdriver for mounting the clamps
Laptop or tablet for camera alignment
Please find more detailed information regarding the first setup in our technical documentation.
Where can I find your terms and conditions; what are the regulations for RMA, guarantees, etc.?
You can find our General Terms and Conditions on our website. Also our Subscription and Support Terms can be found there. Please let us know via email if you need further documents and information.
What are the electrical and/or building requirements to mount the system?
You can find all electrical and building requirements to mount the different Perception Boxes in our product datasheet. Please also have a look in the quick start guide, where the setup is explained step by step. If you encounter issues in this process, have a look at our troubleshooting guidelines. Please feel free to contact our support if you need further assistance.
5. Data Protection
How do you ensure GDPR compliance?
No videos are stored. In addition, the system only sends data, so that active attacks from the outside are prevented. In addition, while the AI collects information about the objects detected in the camera images, it does not collect biometric data or video footage. More information is collected in our GDPR guidelines.
Are the camera images stored? What is stored and where?
Only the data of the configured events are stored. Events can be configured around traffic counts (motorized and non-motorized traffic), origin-destination analysis, and information about objects in a given zone. It is also possible to specify additional parameters that will be included in the event output. Examples are: Speeds, number plates in parking lots for parking time analysis and the maximum number of objects in a zone for a utilization analysis. The transmission of the event data from the Perception Box to the cloud takes place via JSON format to an MQTT broker. More information can be found in our GDPR guidelines.
What biometric data is collected?
We take data protection extremely seriously. The algorithms of our technology are intentionally developed in such a way that no biometric characteristics are collected and thus no persons can be identified. That is also why our technology does not use facial recognition at all.
Do I need a data protection approval or a data protection impact assessment?
No. There is no data collection permit or reporting system required. As long as the data is not linked to other data sources, there is no need for a data protection authorization or a data protection impact assessment necessary (see DSFA-AV). The image generated by the camera exists only about 50 milliseconds and is neither saved nor forwarded. The output of the software is textual data, which is then visualized in the dashboard of the user. Therefore, there is no personal data collected.
Is it possible to "hack" the system?
The system only sends, which means it is invisible to an active attack. It is also not possible to connect to the camera.
6. Offering and Pricing
If the SCC data subscription is terminated, can I keep the data of the subscription period for 3 years?
No. The data will be stored and made accessible for 3 years from their generation date, if an SCC data subscription is active. We strongly recommend downloading your data and transferring it to a different destination, before you end your SCC data subscription.
What’s the difference between the two data subscriptions?
The main difference lies in the data retention. While our standard model (SWARM Perception Subscription) offers a data retention period of 30 days, this can be extended to three years with the SWARM Data Subscription. Further details can be found on our website.
What are the typical costs of a system?
The final costs depend, of course, strongly on the scope and timeframe of the project. You can find our pricing model with all cost factors on our website, as well as a project example with sample costs.
Could we get your system for free for testing?
Needless to say, we will provide you with support for all installations and tests, and we are also happy to send hardware for testing purposes. However, we ask for your understanding that we cannot provide this free of charge and that the expenses have to be covered. Please feel free to contact our Sales team for further details.
Last updated