Most of my projects are developed on S5P6818 development board, which is based on ARM Cortex-A53 core and supports embedded Linux system.
The programming languages I used are mostly C and C++. I also have knowledge of MATLAB, Qt Creator, Keil MDK, etc.
I also have interests in the advanced areas such as machine learning, computer vision and Internet-of-Things.
Some typical projects I had been working are:
Benchmarking Deep Learning on Wearable Platform
Designed and implemented deep learning models on STM32F7 Cortex-M7 microcontroller.
The project focus on handwritten text recognition.
Used TensorFlow framework and Python to implement and train the deep learning models.
The effect of different parameters (e.g. kernel size, learning rate, network layers) were evaluated through experiments and comparisons.
Model architecture and training process were visualized using TensorBoard, providing theoretical support for model optimization.
Memory footprint of the model was reduced by 75% using quantization method, while keeping accuracy over 90%.
The optimized model was obtained via experiments and evaluations, realizing the miniaturization of the DNN model, and its accuracy was maintained.
Vision-based Target Tracking Robot
Designed a robot which is able to follow the motion of a object based on vision system.
Developed a desktop application on PC using C++ and Qt library, achieving the real time
communication between robot and host device based on UDP protocol.
Selected Cortex-A53 SoC as the CPU of robot system. Ported driver of Ralink RT5370 Wi-Fi
adapter and OpenCV library on embedded Linux platform
Using CAMshift as the base algorithm, combined with the Kalman filter and additional measures, an enhanced
joint tracking algorithm was implemented. The real-time performance
and robustness of the algorithm are enhanced, improving its performance in complex environments.
Designed a distance-estimation algorithm based on pinhole camera geometry, realizing realtime distance
estimation function for ground objects, with an error rate lower than 5% within the effective range (3m).
Combined PID controller with Bang-bang controller to controll the real-time speed and steering of robot, keeping
a certain distance between robot and target..
Completed an academic paper as the first author. The paper was presented on the 10th
IEEE Mediterranean Conference on Embedded Computing (IEEE MECO 2021)
[Avaliable at: 10.1109/MECO52532.2021.9460198].
Dual-wheel Self-balancing Robot
Developed self-balancing robot using STM32 microprocessor as main controller.
MPU6050 gyroscope was used to measure real-time attitude and calculate the quaternion through DMP, interfacing with the microprocessor via I2C.
SQLite was used to store user names and passwords, fulfilling the function of log in and
registration.
OLED module was used to display information, interfacing with the microprocessor via SPI.
Implemented 2 nested loop PID algorithm to keep balance of robot and control movements.
Network Communication Application based on TCP
Developed the application in C++ via Qt Creator. Users can make one-to-one communication
with each other.
This project consists two types of applications, the server and the client, based on Transmission Control Protocol (TCP).
SQLite was used to store user names and passwords, fulfilling the function of log in and
registration.
The client set up connection with with server and the server is responsible for message processing and redirection.
An virtual keyboard was implemented in order to input information through touch screen on embedded devices.
Implementation and evaluation of ensemble Kalman filters
Designed and implemented the Ensemble Kalman Filter (EnKF) algorithm based on the theory of standard Kalman Filter (KF).
Simulated the EnKF and KF on tracking problem in MATLAB and quantified the performance by the Root Mean Squared Error (RMSE).
Compared the EnKF and Extended Kalman Filter in a non-linear benchmark problem.
Implemented a 40-dimensional Lorenz model using Runge-Kutta method (RK4) and conducted simulations.
Evaluated the effect of covariance inflation and localization on the EnKF in extremely non-linear system.
Independently finished a 8000-word thesis with grade of 65%.