Education
- Ph.D in Electrical & Computer Engineering, The Ohio State University, January 2020 - December 2023
- Dissertation title: Geopositioning Multiple Autonomous Platforms using Deep Learning and Photogrammetry
- Advisor: Professor Alper Yilmaz
- M.S. in Electrical & Computer Engineering, The Ohio State University, August 2017 - May 2019
- Advisor: Professor Wladimiro Villarroel
- B.E. (Bachelor of Engineering) in Electronic Information Engineering, University of Electronic Science and Technology of China, September 2014 - June 2017
Research Projects
- Autonomous positioning and tracking of aerial vehicles using Geographic AI [
project page] [paper] [link] [video]- Reduced UAS reliance on Global Positioning System, which can be jammed and suffer from multipath problems
- Proposed a real-time pipeline alternative to the GPS functionality using UAS embedded vision system
- Constructed the pipeline consisting of an off-line fast geospatial Quadtree data retrieval, a selection of feature detection and matching schemes, and attitude-control mechanism guaranteeing the scalability of flight region
- Achieved realtime UAS geotracking accuracy by 3.39 meters in average and 5.38 meters in maximum
- UAS Navigation in Real World using Reinforcement Learning (RL) [
paper] [link]- Inspired by human’s instinct: environment understanding and landmark recognition, the team enabled the UAS self-navigation via recognizing its surrounding environment using its embedded vision sensor
- Trained the agent in reinforcement learning framework to interactively learn the navigation policy and familiarize itself using images from vision sensor in an designed UASNAV environment
- Let the UAS fly in the real world to recognize the landmarks and take action according to the learned policy
- Proposed a novel end-to-end UAS navigation framework for long-range vision based navigation in the real world. Experiments demonstrated that the UAS can navigate itself to the destination hundreds meters away from a random selected starting point with following the shortest path
- Object Detection and Height Estimation using Deep Learning [
paper] [video]- Developed object height estimation algorithm using vehicle-mounted monocular vision system and deep learning
- Proposed an end-to-end pipeline consisting of the choice from a selection of advanced object detectors and photogrammetry module doing depth estimation, 2D to 3D backprojection and object height decomposition
- Proposed MOHE-Net detecting and estimating object (over 80 classes) height. Estimated a 183 centimeters tall person within an average error of 5.09 centimeters, around 2.8%
- Artificial Intelligence Technique for Trajectory Estimation of Maritime Vessel [
paper] [link] [video]- Generated georeferenced tracks of maritime vessel traffic based on the data recorded from a single electro-optical camera imaging the traffic from a moving platform
- Localized target vessel among several similar vessels in image coordinates with tracking by detection strategy
- Defined the geometric relation between GPS-based spherical coordinates of latitude and longitude in the world frame, the local camera centered coordinates and the local image coordinates
- Geotracked target vessel in video and retrieved its geoposition in the accuracy by 15.26 meters in average
- Video Background and Foreground Segmentation using Machine Learning [
paper] [video]- Converted moving camera object detection task to foreground segmentation. Segmented objects not belonging to changing background in video setting
- Trained a developed neural network in an end-to-end manner and introduced Conditional Random Fields as a temporal regularization by modeling interactions between previous frames and current CNNs output
- Applied Focal Loss to assign all training samples dynamic weights to enforce more focus on hard examples
- Proposed MBS-Net that achieves 97.53\% Mean IoU on background and 76.06\% on foreground on ApolloScape Datasets
Team Management Experience
- Vision-based Unmanned Aerial System (UAS) Geopositioning & Tracking
- Led junior researcher converting UAS geopositioning pipeline from Python to C++ and deployed it to Jetson Orin. Accelerated inference speed from 15 fps to 35 fps, 133% faster
- Provided guidance for building up virtual environment with Unity game engine. Demonstrated UAS geopositioning pipeline to be working in virtual environment in an online manner
- Real World UAS Navigation using RL
- Partnered with graduated research team to develop UAS navigation in real world using Reinforcement Learning
- Proposed a novel end-to-end UAS navigation framework for long-range vision based navigation in the real world
- Managed UAS data collection and flight results analysis. Demonstrated that the UAS can navigate itself to the destination hundreds meters away from a random selected starting point with following the shortest path
Skills & Qualifications
- Areas of Research: Deep Learning, Computer Vision, Machine Learning
- Programming Languages: Python, Matlab, Git
- Frameworks: PyTorch, OpenCV, Numpy, matplotlib, Pandas, SciPy, scikit-learn, Docker, ArcGis pro
- Coursework: Photogrammetric Computer Vision, Pattern Recognition, VideoGrammetry, Optimization, Image Processing, Algorithms, Intro Time Series Analysis, Project Management
Patent
- System and Method for Hypersonic Aerial Platform Geopositioning (in review)
Publications
- JOURNAL
- CONFERENCE
- A Gis Aided Approach for Geolocalizing an Unmanned Aerial System Using Deep Learning
- Jianli Wei, Deniz Karakay, Alper Yilmaz, IEEE SENSORS 2022
- UAS Navigation in the Real World Using Visual Observation
- Yuci Han, Jianli Wei, Alper Yilmaz, IEEE SENSORS 2022
- DeepTracks: Geopositioning Maritime Vehicles in Video Acquired from a Moving Platform
- Jianli Wei, Guanyu Xu, Alper Yilmaz, IEEE SENSORS 2021
- MOHE-Net: Monocular object height estimation network using deep learning and scene geometry
- Jianli Wei, Jinwei Jiang, Alper Yilmaz, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 2021
- MBS-Net: A moving-camera background subtraction network for autonomous driving
- Jianli Wei, Jinwei Jiang, Alper Yilmaz, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 2021
- Pedestrian localization on topological maps with neural machine translation network
- Jianli Wei, M Taha Koroglu, Bing Zha, Alper Yilmaz, IEEE SENSORS 2019
