| Summary: | 碩士 === 國立成功大學 === 電機工程學系專班 === 101 === An image recognition system usually needs huge resource and is composed of complex parts. These parts include: plenty of memory components, CPU/GPU with rapid operation capability, image capture device or camera, and display devices/interfaces. Recently most of image recognition systems are based on computer system due to modern personal computer (PC) technology that could handle the complex image calculations generally. However, for some special case such as the delivery system of mouth-controlled device (a pacifier switch) for people with disabilities, pure image recognition function is required to operate only for a few minutes during the delivery process, and then the delivery system could be turned off or standby. If the pacifier delivery system is a PC-based image recognition system and it always wait for a long period just to finish only for a few minutes of special task should not meet the requirements of power consumption issue. Therefore, developing a low power portable image recognition system for a special task is a great urgency issue during the time of the depletion of natural resources.
One of the methods to design a low power image recognition system is to integrate the embedded digital logic circuits hardware with the microprocessor, replacing PC to complete the image recognition process and control some special devices such as a robot arm. The embedded image recognition system is in standby mode for most of the time, the low power is needed only for the sensors in the meantime. The system will enter the operating mode to execute the image recognition task while it receives the activation signal from the triggered sensor. Once the procedure of image recognition and control task is finished, the system will enter the standby mode again to save the power consumption.
This thesis reveals a method of image recognition algorithm based on FPGA (Field-Programmable Gate Array), combining with an 8051 microprocessor. A programmable color tone table is defined and adopted in this thesis to make the system suitable for different environments/luminance by editing suitable tone table. It is not necessary to loosen the detection criterion to fit all kinds of skin color for face recognition. The programmable color tone table also can be applied to more complex of multiple object recognition by changing default values in the tables. The portable embedded pacifier delivery system we developed integrated image recognition techniques with robot arm control skills and has been proved that it can function successfully. The accuracy of the system is above 98% when it was tested clinically for 4 normal subjects (two adults and two children). Furthermore the accuracy of test for the subject with severe spinal cord injury is also above 93%. Obviously, the pacifier delivery system is complete in functionality and practicability. We expect that the people with disabilities can interact with more assistive devices with the help of the image recognition system combining with more humanity in the future.
|
|---|
