Embedded Vision Challenges

Abstract

The processing power of modern smart cameras allows more than just simple pixel manipulations for machine vision and inspection tasks. Smart cameras can run complex vision algorithms and thus gradually move the processing and analysis of video streams from large centralized servers to processing ‘on the edge’. This chapter deals with the challenges of bringing high-level vision software from PCs to the limited and constrained development environments of smart cameras. We show that software development on smart cameras can be nearly as comfortable as on a regular PC platform and that processing speeds are sufficiently high for real-time analysis. We use two examples to illustrate how the actual porting of software to a smart camera is achieved. Finally, by employing a very complex pedestrian tracking algorithm pedestrian tracking as demonstration, we highlight the practical challenges of porting a large software system to a device with limited computing resources.