A Pattern Recognition Framework for Embedded Systems
Embedded systems often implement behavior for common application domains, such as the control systems domain or the signal processing domain. An increasingly common domain is pattern recognition, such as determining which kind of fruit is passing on a conveyor belt. Embedded system students typically are not experts in such domains and could benefit from simpler platforms to help them gain insight into the problem of pattern recognition and help them develop such algorithms rapidly. Generic frameworks, such as PID (proportional-integral- derivative) for control, or FIR (finite impulse response) for signal filtering, empower non-expert embedded system designers to quickly build robust systems in those domains. We introduce a generic pattern recognition framework, useful for education as well as for various real systems. The framework divides the task into three phases: feature extraction, classification, and actuation (FCA). We provide template code (in C) that a student or designer can modify for their own specific application. We show that the FCA pattern recognition framework can readily be adapted for various pattern recognition applications, like recognizing box sizes, fruit type, mug type, or detecting vending machine vandalism, requiring only 2-3 hours to create each new application. We report results of a randomized controlled study with 66 students in an intermediate embedded systems class, showing that the framework could be learned in tens of minutes and yielding applications with higher recognition accuracy of 71% for pattern recognition vs. 57% without the framework (p- value=0.03).
- There are currently no refbacks.
1818 N Street N.W. Suite 600, Washington DC 20036