Enhancing the Depth Discontinuities by Pixel-to-Pixel Algorithm Using MPI Thanathip Limna and Pichaya Tandayya Department of Computer Engineering, Faculty of Engineering, Prince of Songkla University, Hat Yai, Songkhla, 90112 Thailand Email: boatkrap@gmail.com, pichaya@coe.psu.ac.th Abstract Stereo vision, especially the Feature-based Stereo Matching (FSM) technique, has been used to enable the visually impaired people to detect obstacles with feature surface. However, it cannot help the visually impaired to detect objects with featureless surface. The Intensity-based Stereo Matching (ISM) technique can detect objects in with featureless surface but it takes longer computing time to process and cannot be used in real-time situation. This paper presents an investigation on the enhancement of the Depth Discontinuities by Pixel-to-Pixel algorithm using parallel computing on a 2-core personal computer and an 8-core server. The results show that parallel computing using Message Passing Interface (MPI) significantly reduces the response time and it is possible to use the ISM technique in real-time. Key Words: MPI, Parallel computing, Depth Discontinuities by Pixel-to-Pixel, ISM, Multi-core 1. Introduction There are many ideas to help the visually impaired with navigation. Electro Neural Vision System (ENVS) [1] is an application of stereo vision for the visually impaired by presenting the obstacles and distances via different signals alerting at their ten fingers. If the object is close, a signal with a high frequency will be used. Likewise, if the object is far away, a signal with a lower frequency will be sent. In this way, the visually impaired can walk to places and avoid obstacles by themselves. However, the technique of the Feature-based Stereo Matching (FSM) [2] used in the ENVS, although can be quickly processed, cannot detect objects with featureless surface which usually cause dangers to the visually impaired. On the other hand, the Intensity-based Stereo Matching (ISM) [2] techniques can detect objects with featureless surface and else. However, it takes much longer computing time and requires running on a computer with high specification. Personal computers nowadays have higher specification than before with acceptably less expensive prices. The market trend is now for 2-core and 4-core Central Processing Units (CPUs) and prices are getting cheaper. Sequential processing cannot fully utilize the maximum performance of multi-core CPUs. Parallel processing techniques are used for enhancing the performance of applications running on multi- core CPUs. So far, there are two methods in parallel programming, including using threads and processes. In our system, the safety of the visually impaired is the key. Using threads, the programmer somehow cannot predict the order of tasks to be processed and cannot specify the number of processes at run-time while parallel programming using Message Passing Interface (MPI) [3][4] has a better process management. MPICH which is one of MPI implementations that has the ch_shmem device that is suitable for the Symmetric Multiple Processors (SMPs) architecture [5] that is used in our stereo vision system. It is also quite convenient to change to other architectures. The source code can be recompiled with other device configurations and used on different architectures. Using parallel computing to help reducing processing time in the stereo vision techniques is interesting and optimistic. Most image processing algorithms including stereo vision can be parallelized. There are a few standards and tools in parallel computing. MPI is used in this work as it is widely known, well documented and stable. In this paper, we present a novel investigation on computing time reduction when applying the parallel computing approach to enhance the Depth Discontinuities by Pixel-to-Pixel stereo (P2P) [6][7] algorithm running on a 2-core personal computer and an 8-core server in order to see the example performances of current PCs and future computers.