19. December 2017

FEV ADAS Annotation Tool enhances development of smart vehicle systems

With the target of assisting its customers in meeting SAE Level 5 Autonomous Driving standards, FEV continues to refine its technologies with the goal of providing a comprehensive suite of tested products that can be easily integrated into the next smart vehicle. The latest example to make development easier is its Advanced Driver Assistance Systems (ADAS) Annotation Tool, which will be demonstrated in the FEV Hospitality Suite at the Bellagio Hotel in Las Vegas during the upcoming Consumer Electronics Show (CES) 2018.

Using a proprietary tool chain, FEV’s ADAS Annotation Tool processes data that has been manually identified, or labeled, through use of a movie. The information would then be continually fed into a vehicle’s Deep Neural Network (DNN), which “learns” to assign a label to certain objects, vehicles, pedestrians, sign, et.al., via an auto-classifier. As the Annotation Tool database grows, so does the “knowledge” of the DNN, and its ability to identify, through the auto-classifier, an increasing number of labels. This provides autonomous vehicle software developers with a cost effective, and time saving tool, since fewer man-hours are required to build the label database.

Labeling is critical to smart vehicle development; therefore, FEV has created a proprietary labeling tool that contains a predefined label collection, including objects typically located on streets like traffic signs, pedestrians, other vehicles, etc. It provides auto-tracking of marked labels, and auto-labeling, looking toward Artificial Intelligence (AI) as a part of the future of automated driving.

The Annotation Tool supports deep learning of neural networks by integrating the movies’ database. Using high-powered PCs for deep learning, it also includes NVidia GPU acceleration. It benchmarks neural networks on demand on selected movies, and collects corresponding statistics.

Like many FEV-developed tools, the Annotation Tool is user-friendly, using a neural network tool chain that is compatible with NVidia GPU acceleration. It is Windows and Linux compatible, and its open architecture allows for further development.