FYP Reflection 2
For this reflection it is about how my project progress and process. A little review on the literature related to the problem addressed. Then, state the scope of working process and the method flow of work. I end this reflection by mention on what knowledge/skills required to do the project and the hardware/software requirement to finish the project.
Date: 16/12/21 (Week 9)
Supervisor Name: Assoc. Prof. Ts. Dr. Nasrul Humaimi Bin Mahmood
Proposed Topic: Assistive Spectacle Camera Recognition of General Object for Visual Impairment Person
Object Detection, Tracking and Recognition for Multiple Smart Cameras
For cameras that have no overlapping fields of views, object association can be achieved by learning the relationship between patterns of entry and exit in various camera views. For cameras with overlapping fields of views, an important issue that arises is fusion of object location estimates.
Wearable Smart Glass: Features, Applications, Current Progress and Challenges
As the smart glass involves many components and instruments, electronic and computer network interfacing devices, the wearer would feel difficulty. It seems that wearable smart glass has good scope for the development as well as the contributions in many application areas.
Low Cost Ultrasonic Smart Glasses for Blind
Traditional navigation device mostly blind cane, blind by tapping the ground or walking around the object to determine the direction, the structure is simple, single function, easy to use, but ordinary cane cannot be proven accurate, such a serious impact on the safety of blind travelers.
Smart Glasses using Deep Learning and Stereo Camera
In the past, the blind people used ordinary cane or trained guide dog. Conventional assistive canes have a low detection rate for obstacles and limited detection area. In the case of guide dog, there are places that cannot be accessed, and there is a high cost for managing and training guide dog.
Deep learning algorithm cannot be performed on low level MCU of smart glasses since this algorithm compute with a lot of data. To perform fast computation, image data is transmitted to the server using wireless communication.
Real Time Multi Object Detection for Blind Using Single Shot Multibox Detector
There are three object detection methods in deep learning based object detection—RCNN, YOLO and SSD. SSD is much faster compared to YOLO and also as accurate as slower techniques that perform region proposals are pooling. So it is suitable for resource constrained devices regarding size, memory and graphical processing speed.
A Convolutional Neural Network based Live Object Recognition System as Blind Aid
The project will accurately detect the surrounding objects using YOLO Algorithm and convert into voice feedback using gTTs. This project made with the help of Deep Leaning and Raspberry pi will greatly help visually impaired individuals to the great extent by acting as a tool connecting them to the world and surpassing their disability of vision.
- Conceptual design/prototype
- Hardware and software development
- Assemble and programming
- Test device and fine tune
- Evolution of the report
- Object
- Real time video
- Microprocessor (Raspberry pi)
- Digital image processing system
- Object detection algorithm
- Deep neural network processing unit
- Object identified images
- Text speech processing unit (gTTs)
- Voice feedback
- Programming skills and understanding
- Testing the programming coding
- Data collection and analyze
- Assemble devices
- Troubleshoot error
- Camera
- Raspberry pi
- Earphones
- Python
- Yolo v3/Single shot detector
- Voice (google text to speech)