The current setup for the Embedded Systems Laboratory course consists of a Gumstix and an FPGA that process a video stream from a single camera. However, the Gumstix is powerful enough to work with two cameras at the same time while it also offers a hardware interface that can support more than one camera. Therefore, it seems obvious that the Gumstix can be used to extract depth information from two camera streams. The objective of this project is to implement so-called stereo vision on the Gumstix and measure its performance. This includes usage of resources as well as frames per second in real-time and the range and resolution of disparity. From that, an estimate can be made about suitable applications for stereo vision on Gumstix.
In order to pursue this objective, some research has to be done in various directions. Appropriate algorithms have to be found that perform rectification and calculate a disparity map between two pictures. The Gumstix camera interface needs to be connected to two cameras. In order to improve performance, the possibility to use the DMA controller or the DSP needs to be explored. As this project builds on Mattanja Venema’s work there is already a camera setup but there many more cameras of similar type that may be more suitable for this project. Additionally, the FPGA can be harnessed to relieve the Gumstix from parts of the workload.