What is Re-ID? Multi-Camera-Tracking in People Counting

Why is Re-ID relevant?

People counting with video analytics is established across numerous industries. In public transport, for example, it replaces manual passenger surveys with continuous, automated counting. For the retail sector, capturing footfall is equally vital. However, a common challenge arises: individuals moving through the fields of view of multiple cameras are often counted multiple times. This is where the new recognition technology known as Re-ID (Re-Identification) comes in. Re-ID recognizes individuals in a privacy-compliant manner, even across isolated camera perspectives, ensuring they are counted only once.

What is Re-ID?

Re-Identification (Re-ID) is an advanced computer vision technology for people counting. It enables the consistent identification and tracking of the same person across different cameras or non-overlapping surveillance areas. Unlike simple people counting, which merely records how many people enter a specific zone, Re-ID offers deeper analytical depth. It utilizes non-biometric metadata to uniquely recognize a person.

What are the advantages of Re-ID compared to classic people counting?

The primary advantage of Re-ID is its superior accuracy compared to pure line-crossing or other counting methods. The same individual is reliably counted only once. Furthermore, Re-ID allows for the differentiation of various groups: one application is distinguishing between passers-by looking through a shop window and customers who actually enter the store. Re-ID can also automatically exclude employees from the count. The algorithms recognize staff by their recurring appearance in the video over long periods, specific uniforms, or typical behavioral patterns. The software tags these individuals as employees and excludes them from the statistics without the need for additional tags or lanyards. Re-ID also offers a clear lead in several other aspects:

Where is Re-ID used?

There are countless cross-industry use cases where people counting with Re-ID is beneficial. Typical locations include:

• Retail stores
• Commercial real estate, such as shopping malls
• Airport terminals
• Train stations and public transport vehicles
• Large logistics hubs, such as ports
• Museums and cultural institutions

What can Re-ID measure?

Re-ID enables the capture of additional KPIs that go far beyond simple entry and exit counts. This opens up complex application possibilities for sectors like retail, transport, aviation, and logistics.

• Dwell Time:
  How long do people stay in the store, a specific department, a vehicle, or a terminal?
  How long does it take to transfer from a train to a bus?
  Is the person a passer-by or a visitor?
• Engagement:
  Do customers interact with each other or with the staff?
• Conversion:
  What percentage of walk-in traffic actually makes a purchase?
  Are there differences between customer segments?
  How likely is a purchase if the dwell time exceeds 5 minutes?
• Group Size:
  Do couples or families separate after entering a store?
  Are groups more likely to buy than individuals?
• Journey Analysis:
  Do customers who visited department A buy more than those in department B?
  What are the typical paths through the store?
  Which areas are avoided? What are the typical routes for commuters in a city center?
• Demographics:
  What is the age range of the customers?
  What is the gender distribution?
  Do specific age groups buy more frequently?

How does Re-ID work without facial recognition?

Re-ID works without biometrics and is fully GDPR-compliant. The protection of identities is guaranteed at all times. To recognize people without identifying their faces or other personal data, Isarsoft Perception utilizes computer vision and AI. From a technological standpoint, Re-ID works as follows:

1. Detection in the video:

As soon as a person enters the field of view of camera A, Isarsoft Perception detects an object of the "person" category. The algorithms analyze metadata - such as clothing color or items carried - rather than biometric data.‍

2. Conversion into a feature vector:

Using deep learning, Isarsoft Perception generates a "feature vector." This is an abstract string of numbers representing the person's characteristics, e.g., blue backpack, gray jacket, blue pants.‍

3. Matching:

If the person leaves camera A and reappears two minutes later at camera B, step 2 is repeated for the "new person." Isarsoft Perception then compares the new features with the previously stored feature vectors and calculates a similarity score. If they match closely, the software assumes it is the same person. The system is robust enough to recognize individuals even under changing lighting conditions or different viewing angles (e.g., from behind instead of from the side).‍

4. From data to insights:

At the end of the journey, the individual data points from various camera perspectives are linked into a trajectory (a movement path), providing operators with valuable information about the complete customer journey.

How is data protection ensured?

At Isarsoft, we consider data protection and information security to be core functions of our software. Safeguarding identities is more than just a legal requirement for us; it serves as the very foundation of trust with our customers and partners. Internal and external audits, along with our ISO 27001 certification by TÜV Süd, confirm our continuous commitment to providing privacy-compliant video analysis software. To ensure that Re-ID remains compliant with data protection regulations, Isarsoft Perception applies the following principles:

Anonymization:

Video data is immediately and irreversibly anonymized; individuals are displayed as pixelated. Matching occurs only via the anonymous feature vector.

Data minimization:

Live streams are analyzed transiently in the local hardware's RAM and discarded within milliseconds. No images or video files are stored on hard drives. Only the mathematical feature vectors are stored for a maximum of 2 hours.‍

Local processing:

Typically, Isarsoft Perception runs on-premise. Unless explicitly agreed otherwise, Isarsoft has no external access to the system or the data.

Learn more about data privacy and information security and get answers to the most important questions about re-id.

Practical example: Process optimization at Salzburg Airport

The anonymized recording of passenger waiting times in the EES (Entry/Exit System) process via Re-ID allows Salzburg Airport to measure and understand process runtimes in detail. The data is used to evaluate workflows in the arrival area. By using existing camera infrastructure, implementation effort was kept to a minimum.

The limits of Re-ID

While Re-ID can reliably recognize people even in crowded environments like station halls, it reaches its limits if an individual's appearance changes significantly, e.g., if a customer changes clothes in a store. Consequently, long-term recognition over repeated visits, like returning to a supermarket a week later, is not possible.

The future of people counting with Re-ID

Re-ID is the evolution of traditional people counting. It offers high accuracy and privacy compliance, making it relevant for operators. This forward-looking technology is expected to replace conventional footfall measurement at entries and exits soon. Furthermore, its scaling potential via integration with IoT systems, Smart Buildings, and innovative mobility concepts opens up new opportunities.

Discover the possibilities of people counting with Re-ID for your industry and learn how Isarsoft Perception can optimize your operational processes.

‍