Multicore DSP

20Stereo vision implementation

CHAPTER MENU

1. 20.1 Introduction
2. 20.2 Algorithm for performing depth calculation
3. 20.3 Cost functions
4. 20.4 Implementation
   1. 20.4.1 Laboratory experiment
      1. 20.4.1.1 SAD implementation
      2. 20.4.1.2 NCC implementation
      3. 20.4.1.3 ZNCC implementation
5. 20.5 Conclusion
6. References

20.1 Introduction

Moving from 2D to 3D representation of an object or a scene can offer an unparalleled advantage since it provides depth information. Stereo vision systems work in the same way as the human vision system which is composed of two eyes and a brain (two cameras and a processor). However, these systems cannot match the human vision system and therefore need to be limited to specific applications to improve the performance. Extracting depth information is a wide research topic that has been around for decades; the number of applications ranges from machine vision, robotics, autonomous vehicles and medicine, and now it is gaining more popularity in embedded and mobile environments due to the advance in low‐cost miniature cameras and high‐speed processes. An example is Advanced Driver Assistance Systems (ADAS) which have a big share of the ever‐expanding automotive electronics market, and this has led to chip manufacturers like Texas Instruments developing the TDA2x SoC [1] and Freescale developing the S32V ADAS MCU that are designed to perform real‐time video analytics, including stereo vision.

The principle behind stereo vision for depth calculation is fairly simple. ...