Data Privacy in Traffic Monitoring: How Collection Technologies Compare

As cities modernize their infrastructure to support Vision Zero and active travel, the method of data collection determines the level of privacy risk. While legacy technologies like loops offered inherent anonymity due to their inability to "see," the shift toward visual data for multimodal accuracy requires strict protocols regarding data processing and retention.

Data Collection from Video

A variety of approaches exist to gather data from video feeds, with materially different privacy standards based on where and how the video is processed.

• Manual Counting: Traditional manual surveys are typically carried out by installing a video camera, which records video for a few days, before that video footage is manually counted either in the US or offshore. This method involves transmitting a lot of Personally Identifiable Information (PII), and exposing it directly to manual analysis.

• Computer Vision with Cloud Processing: Some sensors stream live video to a central server for analysis. This creates a risk of transmitting PII over networks and storing raw footage that could be queried for surveillance.

• Computer Vision with Edge Processing: Advanced systems process video locally on the device's hardware. This method is the industry gold standard for privacy, as it extracts anonymous statistical data(e.g., "1 Pedestrian") at the source, reducing the need to transmit or store visual footage.

Viva’s Approach: Privacy by Design

Viva utilizes edge processing to ensure that data collection remains strictly anonymous. As demonstrated in the New York City Department of Transportation (NYC DOT) pilot, the system is designed to measure street activity, not individual identity.

• On-Device Anonymization: Viva sensors process videoframes on the device itself. Once road users are classified (e.g., Cyclist, Bus, Car) and their paths logged, the video frame is discarded nearly instantaneously as part of normal processing.

• No PII Storage: The sensors do not use facial recognition. Under standard traffic monitoring operations, no raw video is stored or transmitted to the cloud, ensuring compliance with strict privacy standards.

• Validation & Safety Features: For specific features like Near Miss detection or accuracy validation, short videoclips may be retained. In these instances, Viva applies immediate blurring to render faces and license plates unrecognizable before the data is viewable by traffic engineers. Access to this data is restricted by the platform, and governed by contractual controls.

• Data Ownership: In standard municipal contracts, the statistical data produced belongs to the City or Agency, ensuring transparency and public ownership of planning insights.

Model Improvement & Training (Opt-in)

To improve system accuracy and train future versions of its AI models, Viva may collect a very limited number of still images from deployed sensors only where the City or Agency has explicitly given permission.

These images are not part of normal traffic monitoring operations. Each image is manually reviewed to identify any personally identifiable information (PII). If any PII is detected, the image is immediately discarded and not retained.

Only images that contain no PII are incorporated into Viva’s internal training datasets. This process is governed contractually and subject to audit, ensuring transparency and accountability.

Comparison: Temporary Video Surveys

A common alternative for short-term data collection is the "Temporary Video Survey," often used for ad-hoc traffic studies. These methods present higher privacy and security risks compared to Viva's edge-processing sensors.

• Raw Footage Storage: Traditional temporary surveys typically record high-definition video onto physical SD cards or hard drives for the duration of the study (often 24–72 hours or longer).

• Manual Handling Risks: This raw footage, containing unblurred faces and license plates, is often physically transported from the roadside or uploaded in bulk to cloud servers. The physical movement of SD cards introduces a risk of data loss or theft.

• Human Review: Unlike Viva’s AI, which automates classification, traditional video surveys often rely on manual enumeration. This means human reviewers watch hours of unblurred footage to count vehicles and pedestrians, creating a direct privacy intrusion.

• Viva’s Temporary Alternative: Viva offers temporary/portable sensors that utilize the same edge-processing AI as the permanent units. This eliminates the need for SD cards, manual video review, or the storage of long-duration HD video files.

Other Relevant Technologies

• Inductive Loops & Radar: These intrusive technologies are inherently private as they detect magnetic flux or radio wave reflection rather than visual images. However, they lack the granularity to classify vulnerable road users or detect near-miss scenarios effectively.