Need autonomous driving training data? ›

Mighty AI Announces New Offerings for Managing Autonomous Vehicle Data

Mighty AI Announces New Offerings for Managing Autonomous Vehicle Data

Mighty Studio data management portal


Suite of new features includes privacy protections, gaze tracking solution, and data management portal  


Today we’re unveiling a suite of new product offerings available in our data management platform for autonomous vehicle perception data. The new features help our customers better manage ground truth datasets for developing machine learning models for autonomous vehicles. They include facial and license plate blurring for privacy, gaze tracking workflows to improve driver safety, and a portal for managing autonomous vehicle data. These enhancements complement Mighty AI’s cloud-based annotation platform for generating ground truth datasets to improve vehicle perception models.


New Privacy Protections Ahead of GDPR

To protect the privacy of individuals who appear in the raw image data companies collect throughout the development cycle of autonomous vehicles, Mighty AI launched a privacy solution, which includes automated facial and license plate blurring. The automated blurring functionality aims to anonymize people and their vehicles, allowing companies to conceal personally identifiable information before using the data in their development cycles. In addition to this new technology, our community of annotators must also sign non-disclosure agreements prior to accessing images to further enhance data privacy.


“Any time you’re dealing with data that contains personal information, privacy is a concern,” says Mighty AI founder and CEO Daryn Nakhuda. “Autonomous vehicles collect massive amounts of real-world data that must be labeled to train perception systems. Automated blurring solutions that aim to anonymize individuals in that data is a big step forward in protecting their privacy.”


Using a combination of in-house and open source datasets, Mighty AI trained a deep neural net that is proprietary and custom-built to detect faces and license plates in AV scenarios. Today, we’re using the model to automatically apply an irreversible blurring process to the faces and license plates it detects.


The automated blurring application and annotator non-disclosure agreements are part of Mighty AI’s commitment to privacy and providing our customers with the features they need as they seek compliance with various privacy mandates. The most pressing of these mandates, which takes effect May 25th, 2018, is the General Data Protection Regulation (GDPR), a new European Union regulation designed to strengthen data protection for individuals.


Gaze Tracking to Improve Driver Safety

Mighty AI also released tools for the development of gaze tracking solutions. By detecting and monitoring a person’s face, expressions, and gaze, computer vision systems can determine a driver’s ability to take over control from an autonomous vehicle. For instance, monitoring a driver’s gaze inside of a vehicle to see if the driver is at risk of falling asleep.


Mighty AI introduced the tools required for our customers to develop these crucial driver safety systems, which are particularly relevant to designing Level 2 and Level 3 autonomous vehicles. The new tooling generates keypoint annotations for labeling dozens of different points on a human face, which customers can then use to develop computer vision systems to track these points and determine driver attentiveness and readiness to take control from an autonomous vehicle.


Simplifying Autonomous Vehicle Data Management

Companies developing perception systems for autonomous vehicles are drowning in data. In fact, Intel’s CEO Brian Krzanich famously stated last year that autonomous vehicles will generate roughly 4,000 gigabytes—or 4 terabytes—of data per day by 2020. To help address this challenge, we rolled out Mighty Studio, a portal for customers to manage their ground truth datasets.


Mighty Studio is the industry’s first tool to perform quality assurance checks of ground truth datasets against objective measures. With features such as filtering by object classes and date ranges, and the ability to view annotations, Mighty Studio simplifies the process of managing the large volumes of data that companies require throughout the autonomous vehicle development lifecycle.