Speeding up execution time of a smart wheelchair command technique using parallel computing

An Human Machine Interface for wheelchair control based on facial expressions is presented. The interface is implemented on embedded system architecture based on ARM processor rather than personal computer, like usual wheelchair command implementation, to reduce energy consumption while maintaining similar computing performance. The command technique code is complex but offers inherent parallelism. To reduce processing time, two parallelism levels are exploited. The first one is instruction parallelism which is applied to a dual core architecture using OpenMP directives. The second level is data parallelism in which an SIMD specific unit is exploited. In both parallelization techniques, a minimum of initial code re-manipulation is required. As processing unit, we choose the Pandaboard-ES platform that includes a dual-core ARM9 and a set of control interfaces. The obtained preliminary experiments demonstrate the effectiveness of this parallelization and conduct to a 40% reduction of processing time against a conventional x86 CPU.