Frequently Asked Questions (FAQ)

Isarsoft Perception video analytics turns any camera into a smart sensor. Supported features include: People Flow Analytics, Traffic Flow Analytics, Crowd Analytics. Isarsoft Perception can capture occupancy, line-counting, trajectories, dwell time, velocities, geolocation, alarms and much more.

Whether the goal is to make better decisions, improve efficiency, increase customer satisfaction or harden security - Isarsoft Perception real-time video analysis adds value for planning, operation and security.

Isarsoft Perception is used by corporations and institutions worldwide in over 1.000 installations.

Yes, in many real-world, dynamic or crowded settings, Video Analytics tends to deliver higher accuracy than traditional sensor-based counting.

Within the application creation the user can choose from different models. The people model supports the detection of persons. The traffic model supports pedestrians, bicycles, motorcycles, cars, vans, busses, trucks and heavy-trucks (10+ categories) for a granular 'modal split'. The transport model supports people, bicycles, luggage, wheelchairs and trains. Beside the road, additional objects like luggage, dangerous objects and animals can be detected depending on the use case.

It typically takes 4-8 weeks from purchase to start of operation. For larger projects we can start with a Proof of Concept or a pilot project.

Isarsoft Perception is available in three different tiers: Starter, Advanced and Enterprise. For larger projects we can start with a Proof of Concept or a pilot project. We would be happy to show you a demo of Isarsoft Perception and talk about your use case. Contact us now.

Isarsoft, as well as our qualified partners and integrators, are glad to support you from conception to implementation and operation. Reach out to us via the contact page.

Isarsoft Perception is compatible with all major security cameras and leading video management systems (VMS). In most cases no new hardware is required.

The number of cameras a single server can process depends on the build-in hardware. A high performance server can process over 100 cameras in parallel, realtime. In a multi-server setup several servers can be fused to share the load.

Fisheye cameras which support on-camera dewarping can be used similar to regular cameras. Depending on the dewarping configuration of the camera, two or four streams are available for further processing.

Thermal cameras are typically used for fire detection, intrusion detection and road traffic analysis. Isarsoft Perception does not yet support thermal cameras. However, Isarsoft Perception delivers good results with regular security cameras at day and night either through available light sources or infrared light emitted by the security camera.

Isarsoft Perception can be installed on rugged edge devices. The data produced by Isarsoft Perception can easily be sent through LTE or 5G and with some limitations also via LoRaWAN.

Isarsoft Perception is built for maximum connectivity. Other systems can seamlessly and securely be connected with Isarsoft Perception through the GraphQL API, the Python SDK or Isarsoft Perception Cloud Connector. A complete overview is available on the integrations page.

Yes, Isarsoft Perception complies with relevant privacy policies such as the GDPR. Additionally Isarsoft is ISO 27001 certified by TÜV Süd, underscoring our commitment to the highest standards of privacy, information security and risk management. The software does not identify people personally; it does not store faces, personal data, or identity. It only uses abstract visual features (like colors, shapes, and movement patterns) to match whether two detections belong to the same individual.

Isarsoft Perception implements state of the art information security measures and has been successfully deployed in corporate IT environments.

Depending on the version, Isarsoft Perception can be installed on cameras, edge and server hardware. Data processing takes place locally and does not require an Internet connection. The collected, anonymous metadata can be transmitted to third-party systems for further processing.

Yes, the software is fully capable of operating in a completely air-gapped or offline environment. This is a core feature for our infrastructure and security-conscious clients.

Isarsoft Perception transforms video data into metadata. The amount of data produced by Isarsoft Perception depends on how many cameras are being processed and which analytical features are activated through the configuration.

No Personal Identification: no direct identification data (faces, names) and no biometric data are processed. No Video Storage (RAM Processing): live video streams are analyzed only fleetingly in the RAM and discarded within fractions of a millisecond; no image or video material is stored on hard drives, only pseudonymized feature vectors for a maximum of 2 hours. Strict Data Minimization: the result of the analysis consists of purely statistical metadata (e.g. aggregated counts and stay durations) sent to the target platform.

Isarsoft Perception processes live video images from the cameras on a local computer using AI in four steps: (1) Object Detection - identifies object category, position and size in individual frames (Bounding Box); (2) Trajectory Collection - records movement over time, camera-bound, ends when the person leaves the field of view; (3) Recognition (Re-ID) - generates a 500-dimensional feature vector on entry and exit to determine length of stay, deleted after recognition or after a maximum of 2 hours; (4) Permanent Storage - only the anonymized, purely statistical metadata (e.g. length of stay) remains and can be stored long-term, as it has no personal reference.

For Re-ID, no direct identification data such as faces or names are processed, nor is any biometric data. The system merely translates fleeting visual characteristics such as the color of clothing or objects carried by the person into an abstract numerical code known as a feature vector. This vector is neither readable by humans nor can it be converted back into an image. This also means that long-term recognition is not possible with Re-ID. If a person changes their jacket, they will no longer be recognized. This vector is considered a short-term pseudonymized feature for a maximum of 2 hours before it is irrevocably deleted. The final results are purely anonymized statistics (e.g., 'the average length of stay is 42 minutes'). Consequently, inferring the real identity of individuals is technically impossible.

Yes, Re-ID works anonymously and privacy compliant. The software does not identify people personally, it does not store faces, personal data, or identity. It only uses abstract visual features (like colors, shapes, and movement patterns) to match whether two detections belong to the same individual. This makes Re-ID privacy-friendly and compliant with regulations such as GDPR when implemented correctly.

Re-ID is an AI technique that recognizes the same person across multiple video frames, cameras, or viewpoints without identifying who they are. It simply ensures that one person is counted once, even if they appear several times. Video data is anonymized and re-id works without face-recognition.