How to Create an Embedded Vision System

Wed. October 31| 8:00 AM - 10:00 AM | 208B

Conference: ESC Minneapolis 2018

Track: ESC Track D: Advanced Technologies

Format: 2-hour

Pass Type: Conference Pass (Paid)

Embedded vision is ubiquitous, being used across a wide range of applications from autonomous driving to the Industrial Internet of Things. Regardless of the end application to create a solution which offers the highest determinism and lowest latency you will be using either a FPGA or Heterogeneous SoC, which combines programmable logic with high-performance processors.

This session will start with a brief introduction to the different image sensors possible — CCD, CMOS, and IR — as well as the different imagers and selection 1D or 2D sensors. We will also introduce additional elements of the sensor which determine its performance in your system; for instance, integration time, shutter mode, sources of noise, and, of course, the dynamic range.

The main element of the talk will focus on how these sensors can be integrated with our FPGA/SoC and the creation of an image processing pipeline. This talk will examine the following aspects of the embedded vision system:

  • Image processing pipeline overview – Input, Processing, Output, Clock, and Timing.
  • Image capture methods – HDMI, MIPI, VoSPI, Parallel & Serial
  • Output and interfacing with displays – HDMI, VGA, MIPI, GigEVision, etc.
  • Importance of AXI Streaming, conversion to and from AXI Streaming
  • Colour space conversion, video mixing, image formatting
  • Video Direct Memory – what is it, why do we use it, how do we use it.
  • Advanced Image processing using OpenCV – High Level Synthesis and SDSoC


Adam Taylor

Adam Taylor


Adiuvo Engineering Training ltd

Role: Speaker