Real-Time Image Processing on FPGA
Implementation of Miscellaneous Edge Detection Algorithms
In this thesis miscellaneous edge detectors were implemented on an embedded system with a FPGA. Edge detection is an important step in the digital image processing hierarchy because the data rate is drastically reduced. Following units like object recognition obtains only relevant data.
The Roberts, Sobel and Laplace two-dimensional filters were implemented by deriving a lter structure from a one-dimensional FIR filter. The operators can be easily modfied so the implementation of other filters (e.g. low-pass) is possible. Also the implementation of a non-linear filter like the median filter is possible with a pipelined comperator network.
It was necessary to buffer lines from the camera to apply the neighbourhood operators. This leads to the bottleneck of the system: the SRAM access, but in this thesis it was proved that there will be no conflicts in the scheduling.
All filters have their advantages and disadvantages. The Roberts and Laplace filter are easy to implement but they are not resistant to noise. The Sobel filter suppresses noise and extracts the edges well. This is a good compromise which leads to useful results. The Sobel filter is often used in industrial applications.
- Edge detection on FPGA
- Sobel, Roberts, Laplace edge detectors work concurrently
- Hardware: FPGA board Euterpe (Matthias Alles) and Omnivision OV7620 Camera
- Exhibition: Building 11 University of Kaiserslautern
- Download this thesis as PDF here