Design And Develop Object Detection System For Blind People Based On CNN Image Recognition

Lai, Jun Ann (2018) Design And Develop Object Detection System For Blind People Based On CNN Image Recognition. Project Report. UTeM, Melaka, Malaysia. (Submitted)

Text (24 Pages) Design And Develop Object Detection System For Blind People Based On CNN Image Recognition.pdf - Submitted Version Download (366kB)

Abstract

Object detection is a popular topic in visual recognition and plays a significant role in many fields. On the other hand, visually impaired people or blind people are usually unaware of danger that they are facing in their daily life. They faced many difficulties in their activity even in their familiar environments. This project proposes a smart object detection system based on Convolutional Neural Network (CNN) to provide a smart as well as safer living to visually impaired people. The region proposals from edge maps for each image is produced by using edge box algorithm. Then the proposals is passed through a fine-tuned CaffeNet model. The object is detected by the webcam and the feature of the image is extracted, if the object is matching with the trained model in the database which is the cloud storage, then output audio will generate by the system to let the visually impaired people to identify the object. The result is evaluated by using mean average precision (mAP) as well as frame-per-second. As a result, SSD reduced the complexity and archives higher accuracy and faster speed in object detection compared to Fast R-CNN.

Actions (login required)

View Item