An Optimized Vision Library for Embedded Systems
Computer Vision is playing an increasingly dynamic role in equipment such as mobile phones, smart cameras, and automotive safety systems. The signal processing requirements of these applications can be especially challenging: the underlying algorithms are not only computationally expensive, but they also demand robust real-time solutions. One way we found to ease our clients' needs for customized embedded systems was the development of an Embedded Vision Library.
Computer Vision is playing an increasingly dynamic role in equipment such as mobile phones, smart cameras, and automotive safety systems. The signal processing requirements of these applications can be especially challenging: the underlying algorithms are not only computationally expensive, but they also demand robust real-time solutions. One way we found to ease our clients' needs for customized embedded systems was the development of an Embedded Vision Library.
An Optimized Vision Library Approach for Embedded Systems, Goksel Dedeoglu, Branislav Kisacanin, Darnell Moore, Vinay Sharma, and Andrew Miller, Proceedings of the 2011 IEEE Embedded Computer Vision Workshop. (June 2011) Paper Poster
EVA video interview with hands-on demonstration of Lucas-Kanade tracking on TI's C6678 Keystone DSP, where we detect and track thousands of Harris corner features in 1080p HD resolution video. Contributors include Andrew Miller. Full video available, © 2013 EVA.
EVA video interview on various aspects of embedded vision engineering: algorithm design and prototyping in a PC environment, embedded software porting and optimization, and selection of a cost-effective hardware platform. Highlighting the need for embedded-friendly APIs, we review performance speedups achievable through compiler options, loop unrolling, and C6x DSP intrinsics. Full video available, © 2013 EVA.
See also the tutorial on "Moving Object Segmentation (MOS) for Security and Surveillance Applications," at the September 2012 Embedded Vision Summit.
EVA video interview on the trade-offs we faced in creating TI's Vision Library VLIB, challenges and best practices for embedded vision system development. (November 2011)