Stereo Vision Algorithms Suited to Constrained FPGA Cameras

摘要: The advent of cheap RGBD active 3D sensors, such as those based on structured light (e.g., the Microsoft Kinect) or time-of-flight technology, has significantly increased interest in computer vision applications depth data that, most cases, enables higher robustness compared to solutions traditional 2D images. Unfortunately, techniques are quite noisy even completely useless outdoor environments (in particular under sunlight). An effective and well-known technique infer suited indoor is passive stereo vision. Nevertheless, despite frequent deployment this technology many research projects since 1960s, often perceived, especially consumer applications, an expensive due its high demanding computation requirements. In paper, we will review a subset state-of-the-art algorithms that have potential fit with basic computing architecture made low-cost field-programmable gate arrays (FPGAs), without additional external devices FIFOs, DDR memories, etc.) excluding USB GigaEthernet communication controller. Compared more complex designs FPGAs coupled memory devices, clear advantages outlined simplified reduced design manufacturing costs well power consumption. Another significant advantage consists better code portability improved respect obsolescence electronic being almost whole self-contained into FPGA logic. On other hand, mapping similar low-power, poses very challenging task only existing appropriately modified constrained platform. believe proposed would make sensors suitable wider class application scenarios not yet fully addressed by technology.