Skip to content
Research and Project Experience
Ford Personal Route Prediction Project
- Project leader. Be responsible for implementing an online-learning system which can learn the real-world driving pattern and route choice for a specific driver.
- Designed a prediction system using Markov chain model to generate the optimal route (with smallest cost) from origin location to destination at the beginning of the trip.
- The system is intended to improve the current navigation system and also make contribution to improving the efficiency of current power management.
Lane Change Prediction using Physiological Signal
- Project leader. Collected driver’s physiological signals, including ECG, GSR and
Respiration signal, from the real-world driving environment.
- Developed the prediction system framework, including signal
pre-processing, signal synchronization, signal selection, feature
extraction and prediction system.
- Used causality analysis and correlation analysis to select the
useful signals from the signal set.
- A time series prediction system was implemented based on Artificial Neural
Network (ANN).
Cloud-based recommendation system prototype design for Big Data
- Course project for ECE 528-Cloud Computing.
- A Hadoop cluster was deployed on four servers rent in the cloud
provided by DigitalOcean.
- An recommendation system was implemented by using Apache Mahout
and Hadoop MapReduce computation framework.
- The system performance was evaluated on a gigabytes of test data. The recommendation result and computational time provided was compared between the distributed cluster and a single node cluster.
Vehicle Speed Prediction using Deep Learning
- Course project for ECE 679-Advanced Intelligent System.
- Designed a driver vehicle speed prediction system using deep
learning methods.
- Two different deep learning frameworks, Deep Belief
Network and Stack Auto-encoder were used to extract the deep feature
from drivers driving data.
- The features learned by deep network were feed into a traditional
Artificial Neural Network to make the prediction.
Intelligent Bookmark Classifier
- Course project for ECE 579-Intelligent System.
- Designed and implemented a Webpage Bookmark Classifier service
using Python and Qt library.
- Several classification algorithms (such as: Decision Tree, SVM,
Naive Bayes, etc) were implemented and the
classification results were compared.
Autonomous Quadcopter Design
- Leader of a four-member senior design team. Designed a quadcopter from scratch, including chassis design,
component selection and flight controller design.
- Designed an preliminary Arduino based flight controller with a 9-DOF IMU board
(including one triple-axis gyroscope, one triple-axis accelerometer
and one triple-axis magnetometer) and
four ultrasonic sensors.
- Implemented data fusion algorithm to filter the sensor data in
order to provide accurate angle and accelerate information of quadcopter.
Honors and Awards
- As the leader of Team ‘Firmament’, we designed a powerful and easy-to-use ground station app called Fire
- This app helps fire departments quickly respond to fire incidents by making it possible to pre-plan flight routes and transmit real-time imaging from the UAV back to the ground station, letting fire departments conduct a qualitative analysis of the scene, so as to improve decision-making for rescue missions.
2014 ION Autonomous Snowplow Competition
- Responsible for designing computer vision algorithm in the team named ‘GeiLi 3.0’.
- A navigation system, containing both camera and LiDAR system, was designed and embedded to a self-designed snowplow to operate the snowplow to remove snow from a designated path.
- Colored corns were used as the landmarks to provide the boundary information of the snow path feed into the computer vision system. LiDAR system was used to avoid the obstacles placed in the center of the snow path.
TI Cup Undergraduate Electronics Design Contest of Hubei Province
- As the team leader of a three-member team to design Frequency compensation circuit.
- The designed compensation circuit is cascaded after a circuit with the specific sensor characteristic in order to expand the -3 dB cutoff frequency of the sensor circuit and control the passband voltage gain fluctuation and the energy of noise of the circuit when no input signal is given.