This paper describes the architecture and hardware implementation of an embedded, low-cost and low-power dense stereo reconstruction system, running at 30 fps at VGA resolution. The processing pipeline includes an initial image rectification stage, a cost generation unit based on the nonparametric census transform, a state-of-the-art Semi-Global cost optimization stage, and a final minimization and noise suppression step. The hardware implementation is based on a Xilinx Zynq TM System-on-Chip, which besides the FPGA provides a physical dual-core ARM CPU, which is exploited for control and to deliver output over the integrated Gigabit Ethernet connection.

Abstract-Environment mapping is one of the most critical tasks in the development of driving assistance systems and stereo vision has been widely used to accomplish it. However, there are very few datasets that allow assessing the performance of a specific method in a real world application. Most datasets have been created in controlled conditions, thus neglecting scenarios that are impossible to reproduce in a laboratory. In this paper, we present the results of the evaluation of three different dense reconstruction algorithm implementations using a number of well-known strategies that represent different tradeoffs in terms of cost, set up time and accuracy. In our tests, we evaluated two variants of the Semi-Global Matching algorithm, and the Efficient Large-Scale Stereo Matching method, as well as different combinations of additional filters in order to assess their influence on the final behavior of the algorithms.