Ensuring Privacy and Compliance: How Mindy Support Anonymized 500,000 Video Frames for Autonomous Vehicle Development

Data Anonymization, Data Annotation, Quality Assurance

Mindy Support successfully anonymized a dataset of 500,000 video frames, enabling the client to remain compliant with GDPR and CCPA regulations while ensuring the data remained valuable for vehicle and road analysis. Using advanced data processing techniques, the team accurately identified and blurred all identifiable license plates and faces within the video frames. This process maintained the integrity of the visual data, preserving critical elements required for autonomous vehicle development and traffic management insights. Mindy Support’s scalable and efficient solution ensured complete compliance with privacy laws, allowing the client to responsibly leverage their data without compromising analytical precision.

To enhance their autonomous vehicle development and traffic management systems, our client in the automotive industry faced the challenge of processing large-scale video data of roads. These videos, captured from various urban and rural environments, contained sensitive information such as visible license plates and human faces. This posed significant privacy and compliance challenges, as the company needed to adhere to stringent privacy regulations like the GDPR and CCPA. Without an effective solution, the data could not be utilized without risking legal and ethical repercussions.  

The client’s objective was to anonymize a dataset of 500,000 video frames, ensuring that all identifiable license plates and faces were blurred or masked. This anonymization process needed to be scalable, capable of handling high volumes of data efficiently, while maintaining accuracy to preserve the integrity of vehicle and road analysis. Additionally, the solution had to align with privacy laws, ensuring compliance without compromising the usability of the anonymized data for critical insights in autonomous driving and traffic management innovation. Through this approach, the company aimed to leverage data responsibly, advancing technology while upholding privacy standards.  

The client entrusted the project to Mindy Support due to our long-standing relationship , based on excellent performance in regard to various facets of past project actualizations. During the course of our mutual cooperation, we developed an understanding of the peculiar needs and expectations of the client by offering precisely what they needed for their success. With experience in data processing, coupled with a commitment to excellence, this made us a trusted partner where the client had confidence in our ability to manage the task efficiently and effectively. This collaboration history and a proven track record of success sealed the deal for anonymizing their sensitive video dataset.

We started off the project by implementing a multi-step process that combined AI-driven tools along with rigid manual review of all annotation processes that we conducted to actualize the project. This includes the application of advanced algorithms for automatically detecting and blurring personally identifiable information like license plates and human faces within the dataset of 500,000 video frames. In order to ensure the highest level of accuracy, our skilled team members performed critical manual reviews against AI outputs to verify certain edge cases that could have been missed by the AI. This combination ensured full compliance with all data privacy regulations, including but not limited to GDPR and CCPA, without sacrificing the quality and usefulness of the anonymized data for use in their autonomous vehicle and traffic management applications.

Our services on this project included: 

• Implementing a Pre-annotation Algorithm – The initial phase of the project leveraged automated detection tools capable of identifying license plates and faces within the video frames. These AI models were trained to recognize patterns across varying conditions, such as different lighting, weather, and angles, enabling the system to detect sensitive areas quickly. This automation significantly sped up the anonymization process, reducing the manual workload.
• Offering a Human Perspective – Given the complexity of real-world video data, the automated system was supplemented with manual review to ensure accuracy. Participants in the team manually checked the frames flagged by the AI for any missed or incorrectly identified PII. This step was essential, as certain environmental factors, such as glare, shadows, or partial obstructions, occasionally prevented the AI from fully recognizing the sensitive information.
• Daily workload and management – To manage the vast number of frames, the dataset was divided into smaller batches, allowing the 30-member team to work in parallel, enhancing the project’s scalability. The team, led by a team leader and supported by a trainer, ensured that each participant followed standardized procedures, optimizing both speed and consistency across the anonymization efforts.
• Quality control – A dedicated QA team conducted thorough checks to maintain high-quality results. Each batch was sampled and reviewed to confirm that no identifiable information remained visible. The QA process was iterative, meaning any flagged frames were corrected and resubmitted for review, ensuring compliance before finalizing the dataset.

• 500,000 video frames annotated 
• Quality score of 99%+