Rule-Based Action Recognition: Performance Comparison between Lua and GDL Script Language

摘要

The main novelty of this paper is presenting the implementation of GDL classifier in pure Lua language. We also present the evaluation of this implementation for motion capture data classification in real-time and compare the execution time with GDL 1.1 implementation. The experiments showed that Lua implementation performance of GDL in most cases is several times slower than managed C# GDL script 1.1. That is due the fact the Lua is a general purpose language while GDL script is optimized only to cooperate with GDL classifier, however GDL Lua implementation presented and evaluated in this paper proved to be fast enough to be useful both in scientific and in commercial applications. The performance time in productions scenarios (classification on relatively "big", 40 kB rules definitions) can be done with frequency approximately 100 executions per seconds which should be enough for up-to-date motion capture solutions. The new GDL Lua Implementation seems to be a good successor of GDLs 1.1.