CELL处理器并行实现立体匹配算法

Cell processor stereo matching parallel computation

提出了适应于IBM CELL BE并行处理器的两种立体匹配算法——动态规划(DP)算法和置信传播(BP)算法的实现方案。改进了BP算法中的数据成本计算方法,采用重叠窗方法保持BP算法原有的匹配效果。实验证明,基于CELL处理器的并行实现极大地提高了生成视差图的运算速度。针对室外场景噪声较大的特点,提出采用Sobel边缘检测和Residual算子对图像进行预处理,实验结果表明,该方法改善了生成视差图的效果。本文算法在低成本硬件设备平台上提升了立体匹配技术的计算效率,在汽车自动和辅助驾驶系统、多媒体视觉技术的硬件一体化中具有重要的应用价值。

This paper presents an efficient parallel implementation of belief propagation and dynamic programming algorithms on an IBM CELL BE processor （Sony Playstation3） that speeds up stereo image matching. Overlapping windows and improved data cost with pyramid computation are adopted for implementation of belief propagation algorithms. Considering that real-world images are typically degraded by various types of noise, differing exposures and changes in light etc, Sobel edge and residual operator are used for image pre-processing to improve the stereo matching results. The proposed method has important application on hardware unification of automatic driving binocular computer vision together with multimedia visual technique.