| Summary: | 碩士 === 國立聯合大學 === 電機工程學系碩士班 === 106 === With the advancement of deep learning technique, the accuracy of image recognition in the field of computer vision has been improved greatly. Many research results can be further implemented into practical products through the use of embedded system development platform with movable characteristics, which can bring the human beings a more convenient, safe and comfortable life. In this thesis, a blind aid system which can support blind people walking during the day and night, is proposed by using both the Jetson TX2 embedded system and deep learning technique. The proposed system mainly uses the deep learning technique to recognize the images which are extracted via the webcam to detect the target object needed for the blind person when he walks. Moreover, the orientation judgment function of the target object is also provided in the proposed system to make the blind person much easily know the locality of the target object. Finally, the proposed system returns the recognition results in speech so that the blind person can know the identification results simply through the earphones. The experimental results show that the proposed system using the YOLOv2 network and Tiny YOLO network can achieve the image recognition rates of 86.09%, and 82.27%, respectively. Although the overall system speed is slightly lower due to hardware limitations, the proposed system still has a good performance in day and night image recognition rate, which can also verify the reliability and feasibility of the proposed system.
|
|---|
