Project

# Title Team Members TA Documents Sponsor
14 Portable Flower Recognition Device
 Qichao Gao
Tianying Zhou
Zekun Wei
 Nicholas Ratajczyk other0.pdf
other0.pdf
proposal0.pdf
#Problem
When outdoor lovers go to experience the wild nature, they may face something quite challenging: they would like to identify a cluster of flowers in from of them, but they lack the knowledge to do so. Even if they take a picture at that instance, they may stay in an out of service area and have trouble figuring out what flower it is without searching on internet.

#Solution Overview
Our solution for clearly recognizing flower species without search engine is a portable flower recognition device with a camera and a display screen. An embedded processor will run trained DNN when receiving image from the camera, and then send the classified result and information about that flower specie to the display module. The whole device would be handheld, with the embedded module and all other modules fitting into a 3D printing case from machine shop.

#Solution Components

##Camera Subsystem
- A camera (OV7725 camera module) will be used to capture images of live flowers
option1. put it on glasses and transfer image through WIFI/bluetooth module.
option2. integrate it with processing unit

##Processing Subsystem
- We plan to train a DNN model with existing flower database from Oxford Flowers dataset FLOWERS17 and FLOWERS102 which contain 17 and 102 flower categories. Once we have the trained model, we will build a embedded processor which has the capacity of doing the forward propagation in a relatively short amount of time.
-The components of our embedded processor will include a STM32 microcontroller, a Micro-USB for power supply, HDMI/VGA for display module and the camera viewfinder.

##Power Subsystem
-A battery module powering the processing subsystem through Micro-USB
option1. lithium battery
option2. dry-cell battery

##User interface Subsystem
- A simple display screen by LCD( ILI9341 module) would give the user feedbacks.
-Controller
option1. make it a watch if feasible
option2. integrated with processing unit

#Criterion for Success
- Successfully capture the flower image with camera on viewfinder screen
- Can classify the species of flowers with a relatively high accuracy (above 60%)
- Accurately run forward propagation on our embedded processor
- Successfully give classification result and provide visualization on screen.

We think this project should be considered appropriate for a senior design project since it requires both software and hardware knowledges. To successfully finishing this project we need to know machine learning knowledge to write the code and train the model and hardware knowledge to design a PCB and integrate all the units on our own. Also, there is no such product on market for campers that can recognize flower without internet. I believe this product can perfectly solve the problem we have .

Idea post url: https://courses.engr.illinois.edu/ece445/pace/view-topic.asp?id=27213

Dynamic Legged Robot

Joseph Byrnes, Kanyon Edvall, Ahsan Qureshi

Featured Project

We plan to create a dynamic robot with one to two legs stabilized in one or two dimensions in order to demonstrate jumping and forward/backward walking. This project will demonstrate the feasibility of inexpensive walking robots and provide the starting point for a novel quadrupedal robot. We will write a hybrid position-force task space controller for each leg. We will use a modified version of the ODrive open source motor controller to control the torque of the joints. The joints will be driven with high torque off-the-shelf brushless DC motors. We will use high precision magnetic encoders such as the AS5048A to read the angles of each joint. The inverse dynamics calculations and system controller will run on a TI F28335 processor.

We feel that this project appropriately brings together knowledge from our previous coursework as well as our extracurricular, research, and professional experiences. It allows each one of us to apply our strengths to an exciting and novel project. We plan to use the legs, software, and simulation that we develop in this class to create a fully functional quadruped in the future and release our work so that others can build off of our project. This project will be very time intensive but we are very passionate about this project and confident that we are up for the challenge.

While dynamically stable quadrupeds exist— Boston Dynamics’ Spot mini, Unitree’s Laikago, Ghost Robotics’ Vision, etc— all of these robots use custom motors and/or proprietary control algorithms which are not conducive to the increase of legged robotics development. With a well documented affordable quadruped platform we believe more engineers will be motivated and able to contribute to development of legged robotics.

More specifics detailed here:

https://courses.engr.illinois.edu/ece445/pace/view-topic.asp?id=30338

Project Videos