Parallel Gesture Recognition with Soft Real-time Guarantees

Student thesis: Master's Thesis


In an effort to improve the quality of interactions between users and computers, interest in multi-touch input, optical gesture recognition, accelerometer-based input, and voice recognition on mainstream, consumer hardware has emerged in recent years. To power these natural user interfaces, a complex correlation has to take place of the primitive sensor readings that are collected by this new generation of input devices. However, detecting patterns using imperative programming languages is cumbersome, error-prone, and lacks flexibility. Declarative techniques are usually put to effect, in the form of rules that are defined in a declarative programming language, and interpreted by inference engines.
To allow the performance of those engines to increase on modern and near-future commodity hardware, they must be able to process the events in parallel. However, engines designed to benefit from multiple processing cores do so at the cost of unpredictable execution times: They do not guarantee to recognize patterns of events in a bounded amount of time. This leads to unresponsive and unpredictable systems, which results in a bad user-experience.
This thesis proposes the Parallel Actor-based ReTe Engine (PARTE), a complex event detection system which uses an actor-based parallel programming paradigm, while offering soft real-time guarantees on the time required to detect the patterns. It is designed to run on modern and near-future commodity hardware, i.e. multicore computers using a shared memory architecture. PARTE does not rely on support from a real-time operating system, nor on exotic synchronization instructions which are not present on off-the-shelf hardware. Our system can be integrated into systems which have a need for a responsive and predictable complex event detection system, such as for instance gesture-recognition systems.
We show that PARTE solves the aforementioned problem of predictability by offering soft real-time guarantees on the detection of events, and that the proposed design results in a responsive, performant system, which is scalable on modern and near-future multicore computer systems.
Date of Award14 Sep 2012
Original languageEnglish
SupervisorWolfgang De Meuter (Promotor), Theo D'Hondt (Jury), Tom Van Cutsem (Jury), Ann Nowe (Jury), Stefan Marr (Advisor) & Lode Hoste (Advisor)


  • Parallel
  • Actor
  • Rete
  • Gesture Recognition
  • Complex Event Processing
  • Soft Real-time

Cite this