Predictive Analytics Times
Predictive Analytics Times
EXCLUSIVE HIGHLIGHTS
SHARE THIS:

1 month ago
Wise Practitioner – Predictive Analytics Interview Series: Miles Porter, Ryan Wolbeck and Josh Chapman at Trimple Transport Mobility

 

By: Luba Gloukhova, Founding Chair, Deep Learning World

In anticipation of their upcoming conference presentation at Deep Learning World Las Vegas, June 16-20, 2019, we asked Miles Porter, Data Scientist, Ryan Wolbeck, Data Scientist, and Josh Chapman, Data Engineer, at Trimple Transport Mobility, a few questions about their work in deep learning. Catch a glimpse of their presentation, Safely Driven: How Trimble Transportation Mobility is Applying Deep Learning to Make the Roads Safer for Everyone, and see what’s in store at the DLW conference in Las Vegas.

Q: In your work with deep learning, what do you model (i.e., what is the dependent variable, the behavior or outcome your models predict)?  

A: We use deep learning to detect objects that may exist in front of class 8 semi-trailer trucks.  Our models run on video captured by the truck.  These videos are triggered by sonor, radar and/or lidar, when those sensors detect an object that may be in the path of the vehicle.  Once our models have classified objects and their locations, secondary models are used to determine if dangerous driving situations exist.  Our models don’t drive the vehicles, rather they are focused on identifying dangerous situations in videos captured by the truck.  Once those situations have been identified, we bring the videos to the attention of fleet safety managers.

Q: How does deep learning deliver value at your organization – what is one specific way in which model outputs actively drive decisions or operations?  

A: Our models triage video captured from semi-trucks and saves our customers time in reviewing the videos.  Some larger fleets can have over 80,000 videos to review in any given month.  A vast majority of those videos don’t include any dangerous situations.  Our modeling system can save a participating trucking fleet hundreds of people hours each month reviewing videos “by hand”.  Furthermore, reviewing the videos can be very tedious and prone to human error.  Our deep learning models never “get tired”, and their accuracy does not decline over time.  We like to think that we deliver value to more than just our customers.  Accurately reviewing safety videos helps makes the roads safer for everyone.  

Q: Can you describe a quantitative result, such as the performance of your model or the ROI of the model deployment initiative?  

A:  Since October 17, 2018, we have reviewed well over 200,000 videos captured from trucks.  Each video takes about 1 minute for a human to review.  Based on feedback from our customers, our system is around 85% accurate at detecting dangerous driving situations with a 2% type II (True Negative) error rate.  In other words, we error on the side of saying that a video shows a safety issue while our customers think it is safe. 

Q: What surprising discovery or insight have you unearthed in your data?    

A: A vast majority of the videos we review show that professional class 8 semi-truck drivers operate their vehicles in a very safe way.  Clearly , they are professionals, and we have hours of video that shows that.  Many times, when dangerous situations do occur, they are caused by individuals that are driving passenger vehicles in unsafe ways.

Q: What excites you most about the field of deep learning today? 

A:  Our systems run in the cloud, after the video has been captured.  We are very excited, and are actively pursuing, techniques to bring our models “to the edge”, where they can be of greater benefit to the driver.  Many trucking companies own tens of thousands of vehicles.  The cost of replacing their vehicles with autonomous versions would be astronomical.  Also, the autonomous driving software, while advancing rapidly, is still not at a point where it can completely replace a professional class-8 semi-driver in every situation.  We are excited about how deep learning can help augment the skills of current drivers and continue to help make the roads safer and the trucking industry more efficient.

Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Deep Learning World.

A:  Using deep learning systems in the real world is more than just creating deep learning models and training them.  Successful delivery of these systems involves getting deep learning models, infrastructure and associated applications up and running in a stable production environment.  The systems must be able to scale, be resilient to failures, able to recover quickly, and they must be maintainable in the long run.  Our presentation will not only explore our deep learning model, but also how we have approached delivering the model within the larger Trimble ecosystem to continuously help ensure the safe operation of class-8 semi-trucks for the benefit of everyone.

—————————–

Don’t miss their presentation, Safely Driven: How Trimble Transportation Mobility is Applying Deep Learning to Make the Roads Safer for Everyone, at DLW on Tuesday, June 18, 2019 from 11:00 to 12:00 AM. Click here to register for attendance.

By: Luba Gloukhova, Founding Chair, Deep Learning World

Leave a Reply