Prof. Dr. Peter Nauth
Embedded Intelligent Systems
Definition and Introduction
Embedded Intelligent Systems deal with Embedded Systems and Intelligent Systems.
Embedded Systems are microcontrollers which are embedded in a device (system, machine, product) and perform a dedicated task for this device as opposed to general purpose microcontroller in a PC. Examples are microcontrollers in elevators or in the ABS (Anti-Blocking-System) of cars.
Intelligent Systems use intelligent algorithms perform complicated tasks such as decision making or autonomous navigating in a complex environment.
Hence, Embedded Intelligent Systems are embedded systems which execute intelligent systems algorithms. Examples are microcontroller in autonomous robots, in or in mobile phones with advanced functions such as speech recognition. The microcontrollers involved must provide interfaces such as A/D converters and PWM (pulse width modulation) for data exchange with sensors and actuators in order to interact with the environment.
The main components of a device controlled by an Embedded Intelligent System are:
• Sensors and Actuators for interaction with the environment • Intelligent Algorithms such as methods for analyzing complex sensor signals, sensor fusion, decision making and action planning for achieving a goal • Microcontrollers for the acquisition of the sensor signals, running the intelligent algorithms and controlling the actuators Autonomous System (e.g. autonomous robot) consisting of many embedded intelligent systems:
Intelligent Algorithms: Pattern Recognition
Definition and Processing Steps
Pattern Recognition can be defined as the
Analysis of a complex signal (= pattern) in order to recognize objects and calculate their properties or to understand the meaning of the pattern in context with a task and environment
In a typical sequence of processing steps (fig.1) the analog signal containing the pattern such as a video signal from a camera is acquired by digitization. The digitized pattern is segmented into objects, Next, parameters characterizing each object are extracted and classified based on a classification model. The result calculated from the pattern is a number of classified objects such as a recognized word (speech recognition), a recognized obstacle (robot vision) or a bad quality product (vision inspection system).
Analog Signal f (a)
Signal Acquisition and Pre-Processing
Digital Signal f (s)
N Features per object
Assignment of each object to a class Ωi
Fig.1: Processing steps of an Pattern Recognition Systems
If the result is regarded as a hypothesis, it requires iterations with hypothesis driven modifications on the processing steps.
Depending on the applications not all of the steps mentioned above might be necessary.
Pattern Representation, Signal Acquisition and Pre-Processing
Analog signals representing patterns may be modelled as a set of functions depending on a number of variables. The functions contain amplitudes, intensities or concentrations of physical or chemical data, the variables are dimensions in space and time usually. Therefore the mathematical representation of a pattern is as follows:
f (a) = [pic]
Some examples of signals containing patterns are:
• Speech signal:
f(a) = p(t)
p(t) is the sound pressure
containing the speech
• Monochromatic Images:
f(a1,a2) = b(x,y)
b(x,y) is the brightness as a function of the spatial...
Please join StudyMode to read the full document