Person detection may be performed using different means. Facial recognition is quite famous, however we may use several methods to detect people with different level of accuracy. 200 words.
Considering recognition of objects, it is important is to understand that detection may be performed using different methods e.g. various spectrums of radio waves. Inside of each physical method, more detailed method may be used e.g. detecting digital signal in given frequency, detecting color, sound frequency, temperature, distance, or making complex image or sound recognition.
Each object has different characteristics, emits different type of energy, moves differently, or even vibrates in unique way.
In object recognition model the detection system should be able to learn by measuring target object by using all range of available means to correlate them and build digital model describing the target object.
The description model should be usable in recognition phase even by being able to use only subset of information. Detected information is matched agains model to be passed to recognition phase, which sends feedback bask to learn module. This feedback loop makes the system self learning. After some rounds of detection by various means, the system is able to answer with known probability the question: which object is currently in range of sensors.
Hmmm..... by various means:
Second step is to apply wights to each of properties or to groups of them. Third is to let detectors collect information. Fourth is to let machine learning engine to correlate information being collected. Fifth is to perform detection having even partial enformation from subset of detectors.
Sounds possible? Certainly yes, but sound like a job for a neural network.
Considering recognition of objects, it is important is to understand that detection may be performed using different methods e.g. various spectrums of radio waves. Inside of each physical method, more detailed method may be used e.g. detecting digital signal in given frequency, detecting color, sound frequency, temperature, distance, or making complex image or sound recognition.
Each object has different characteristics, emits different type of energy, moves differently, or even vibrates in unique way.
In object recognition model the detection system should be able to learn by measuring target object by using all range of available means to correlate them and build digital model describing the target object.
The description model should be usable in recognition phase even by being able to use only subset of information. Detected information is matched agains model to be passed to recognition phase, which sends feedback bask to learn module. This feedback loop makes the system self learning. After some rounds of detection by various means, the system is able to answer with known probability the question: which object is currently in range of sensors.
Car detection
Let consider use case of car detection. First question is how a car may be identified?Hmmm..... by various means:
- by color,
- by heigh, width, and length
- by weight,
- by sound of the engine,
- by temperature of the engine,
- by location of exhaust pipe, and number of them,
- by vibrations,
- by sound of the engine,
- by sound of opening, closing doors,
- by speed,
- by type of lights,
- by location of lights,
- by parking lot where the car is parked,
- by plate number,
- by Bluetooth identifier of on board computer,
- by RFID used to open car park gate,
- by electronic code associated with digital key,
- by date and time when a car was detected,
- by driver.
- by duration of being parked with still running engine.
Second step is to apply wights to each of properties or to groups of them. Third is to let detectors collect information. Fourth is to let machine learning engine to correlate information being collected. Fifth is to perform detection having even partial enformation from subset of detectors.
Sounds possible? Certainly yes, but sound like a job for a neural network.
###
No comments:
Post a Comment