Implementing an Optimized CNN Traffic Sign Recognition Solution

Monday, May 1, 3:45 PM - 4:15 PM
Summit Track: 
Technical Insights I

Now that the benefits of using deep neural networks for image classification are well known, the challenge has shifted to applying these powerful techniques to build practical, cost effective solutions for commercial applications. In this presentation we’ll explain how we designed, implemented and deployed a real-world embedded traffic sign recognition solution on a heterogeneous processor.

We’ll show how we used the TensorFlow framework to train and optimize an efficient neural net classifier within the constraints of a typical embedded processor, and how we designed a modular vision pipeline to support that classifier. We’ll explain how we distributed the whole vision pipeline across the compute elements of the embedded SoC, present our methods for evaluating and optimizing each stage, and summarize the tradeoffs made in the final implementation.

We’ll also touch on solutions to some of the practical challenges of deploying CNN-based vision products, including support for multiple concurrent networks on a single device and efficiently managing remote network model updates.


Rafal Malewski

Head of Graphics Technology Engineering Center, NXP Semiconductor

Rafal Malewski leads the Graphics Technology Engineering Center at NXP Semiconductors; a GPU centric group focused on graphics, compute and vision processing for the i.MX microprocessor family. With 16+ years of experience in embedded graphics and multimedia development, he spans the full vertical across GPU HW architecture, drivers, middleware and application render/processing models. Rafal is also the EEMBC Chair for Heterogeneous Compute Benchmark (HetMark)

Sébastien Taylor

Vision Technology Architect, Au-Zone Technologies

Sébastien Taylor is the embedded Computer Vision technology lead at Au-Zone Technologies.  Bringing over 18 years of experience in embedded software product development, his work focuses on the design, implementation and acceleration of traditional computer vision algorithms and neural network deployments on embedded vision platforms.  Recently Sébastien has been leading an internal team developing Au-Zone’s graph based CNN development IDE (DeepView) and embedded inference engine (DeepZone)

Please plan to join us May 22-24, 2018 in Santa Clara, California.