I have posted several interesting projects. Here is the catalogue:
Graduate:
Undergrad:
(Complteted during my internship at Tio Companies, presented at TechConnect World 2024)
This study leverages high-resolution microCT imaging and computational analysis using Lightweight Analysis of Mouse Anatomy (LAMA) and ShapeWorks to investigate phenotypic variations in E18 mouse fetuses. Significant morphological differences were identified across multiple organs, highlighting the potential of these tools in advancing our understanding of genotype-phenotype relationships. The integration of phenotypic data with genomic and environmental inputs provides a promising pathway for uncovering causal mechanisms behind developmental anomalies.
These findings illustrate the potential of combining LAMA and ShapeWorks in detecting and quantifying phenotypic variations, with implications extending to the broader context of genetic research and developmental biology. Three-dimensional phenotypic analysis combined with genomic, proteomic, environmental, and experimental data with high-throughput computational analysis and causal inference holds the potential to dramatically improve our understanding of complex developmental disorders.
This pipeline, currently hosted on the Craniorate website craniorate.org, is designed to automate and standardize the evaluation of craniofacial features from medical imaging. It streamlines the process of craniofacial analysis from CT scans, allowing for a wide range of conditions such as different poses, coverage areas, patient ages, and stages of bone formation.
The workflow integrates multiple steps into a cohesive pipeline: initially, the CT scans are resampled to uniform resolution to correct any distortions, followed by the basic localization and cropping of the head region. The segmentation step then differentiates between skull, skin and background, filling any anatomical gaps to complete the skull structure. The refined skull is subsequently aligned to a standardized template, with critical landmarks propagated from the template for consistent cropping. Then it conducts smoothing and shape analysis of the cranial structure, ensuring high-quality outputs ready for clinical assessment.
From its inception in 2020 through to 2024, the pipeline has processed over 1,800 CT scans from more than 15 institutions, encompassing a variety of cranial conditions including different types of craniosynostosis—both before and after surgical interventions. Efficiency is paramount, with the system capable of analyzing each scan in just a few minutes when operational on a multicore server. This automated solution not only accelerates the diagnostic process but also democratizes advanced craniofacial analysis, making it accessible to non-specialists.
The DBLP Computer Science Bibliography dataset contains more than 1.2 million bibliographic records. Hence, for researchers, it is a useful tool to trace the academic works and to get biblio-graphic details when composing the list of references for the new papers. While there are a lot of works on cs rankings based on DBLP dataset, we found that it is also interesting to study the collaborations in computer science between universities using this dataset.
The motivation of this project is from a small talk. One of our group member would like to know which university has the most relationships with other institutions in computer science, indicating the willingness of this university to communicate with other institutions over the world. Based on the DBLP dataset, we define such relationship as the number of publications two institutions have worked with together. And we believe that we are not only people who areinterested in such kind of questions, like which university are the most active in computer science and which institution they should go if they are interested in doing research in certain speci c areas in computer science. It is become a natural choice for us to run this project.
Overall, our project is supposed to provide the interactive visualization of worldwide affliations research achievements in top CS conferences. Publication amount as well as research cooperation will be displayed together. We will use group of charts to show the relationships among the univer-sities in the dimensions of location, topic field, and time. According to the data scales, the charts can be divided into three groups: world view, contrast view and certain university view.
We preprocessed the data of cooperation among CS institutions and could discover the phenomenon of clustering of cooperation with respect to distance and area such as AI. Therefore, the answer is that there is correlation between distance, research topic and institution cooperation, while the infuence of distance is more noticeable.
The project could clearly display the cooperation network of the institutions. And the users are provided adequate freedom to select the data to be displayed, by years and institutions, in various forms of scatter plot, line chart, ranking table and etc.
It could be further improved with adding more features to the plots to display more concert data such as information detailed to the level of conference.
Current pose estimation methods have achieved pretty good accuracy. This project applied Alpha-pose and it performed fairly well even in the scenes with crowds of people with high speed.
The angular representation could encode the major characteristics of the poses in 2D. Images are well categorized using it with Kmeans.
This experiment is based on the CyberTORCS platform for intelligent vehicle control simulation experiment. Simulation platform is developed by Prof. Yang Ming and laboratory members Xuangui Huang, Yang Chenxi development. Users can design and write their own algorithms in C++, with the use of vehicle control input and output variables, according to road bending, vehicle yaw and other factors to change the body state, to achieve the vehicle in the platform of automatic travel. The experiment is divided into two parts: follow the car experiment, parking experiments.
Follow-the-car experiment requires the design of the pilot traffic behavior, while writing the car’s following algorithm, so that it can follow the pilot car driving. The car can not collide with the pilot ship, nor can it pass the pilot, the average distance between the two cars is aimed to keep at 20m The criterion of the algorithm is, the smaller the deviation from the vehicle, the smaller the collision volume, the more perfect the algorithm is. Our group win the 1st place in the course’s final-competition among 27 groups, mainly using PID Control.
The parking experiment gives the coordinates of the five different parking spaces, requiring the algorithm to be programmed so that the vehicle can park into the parking space in the form of a vehicle at the rear or at the front. The parking spaces are 5 known parking spaces and one unknown parking space. Accuracy (the deviation of the four vertices of the vehicle from the standard position) as the criterion, the faster the speed, the higher the accuracy, the more perfect the algorithm. We develped several shifting algorithms for parking and above is the most facinating one.
I applied 8253 clock pulse to generate an interrupt signal, 8259 chip to determine the priority, in order to test the four digital tube at the same time display and clock timing.
8259 is mainly used to achieve the interrupt priority to determine the function, so that only one interrupt line 8086 can handle multiple interrupts.This experiment uses more variables, directly on the variable increments, modulus operations, simplifying a number of steps.
We applied the positive kinematics in MATLAB, according to the random connection angle, calculate the three-dimensional coordinates of the generated point, draw the random three-dimensional space points, through a large number of random points to draw the PUMA’s range of motion, and produced a range of three views of the movement.
We applied the inverse kinematics, entering the coordinates of the end of the starting point, to generate the path, calculate the angle of the arm link, through the path point planning, set the middle point, set the middle of the two obstacles above, and find out the path detouring obstacles.
Line Following and Maze Solving for Intelligent Mobile Robot on Wheels track-the-line’s the general idea is: the robot at the bottom of the installation of two photosensitive sensors, by reading the two photosensor readings to determine which side of the robot line or outlet, with the feedback to the timely progress of the robot to adjust the state, So that the robot at all times along the black line.
Due to the limitations of the actual equipment, we use the Bug algorithm. When the collision sensor detects an obstacle in front of it, the robot moves around the obstruction until it is detected that it is returned to the black line again. When the collision sensor detects an obstruction, the controller enters the interrupt.
In the “maze” mission, our goal is to use the sensor to detect the surrounding environment, the design algorithm to control the robot, in the direction of the maze to choose their own direction, the implementation of the action, so that from the maze of one end of the opening, went to the other end Open. The topographic map of the maze is shown below. At the same time, the presentation time is limited, so another limitation of this task is that the time to finish the labyrinth must be within three minutes.
First of all, take the maze of the general idea is: always along the right / left side of the wall to go. By observing the maze of the topological structure, it is not difficult to find that we walk the maze without isolation, not connected with other walls of the wall, so is a “single connected maze”, there is no single loop in the maze, the maze is the correct path segmentation Into two parts, so as long as the side of the wall to go, always able to get out of the maze.
For the choice of hardware sensors, we use infrared sensors and collision sensors to detect the environment in which the robot is located. Infrared sensor responsive, but the measurement range is small, suitable for measuring the distance with the right side of the wall. The collision sensor is also responsive and can be placed separately to facilitate the expansion of the wall can be measured, so it is used to measure the front wall.
First of all, the camera to collect the image of the gray-scale transformation, and then the gray-scale image to enhance the local noise, automatic threshold segmentation to obtain binary images.
The character training set is obtained by training the OCR (opti-cal character recognition) by the machine visual function, positioning the digital area, setting the character spacing, threshold and normalization. Finally, the numbers are compared with the characters in the character training set and the recognition results are displayed.
Road signs or certain logos recognition for car driver is very important and convenient, especially when coming to unfamiliar areas, drivers often ignore the road signs. For example, speed limit sign, one-way street sign, the prohibition of turning sign and so on. Through this system of automatic road signs recogniter, drivers can focus on driving conditions without to pay more attention to other signs and can provide the driver with guidance and security.
With the help of the recognition, drivers may miss fewer destinations and will be reminded of the special rules on the road. The method we use is to find out the gradient variation of local image as the feature points, that is SIFT tranformation.
Baidu Post Bar is a broad information platform, where you can find the same interest through a simple search of the collective, to provide a wealth of information, some need to save a lot, especially some posts may soon be invalid, so I designed a “ Post Crawler “to meet this demand, which will specify the contents of the post and saved as txt and jpg file, making is convenient and concise.
If you are interested in this work or have any questoin, please contact me via wenzheng.tao@utah.edu.
The copyright belongs to the author. For business reprint please contact the author to obtain authorization, non-commercial reprint please indicate the source.
©2017-2020 by Wenzheng Tao. All rights reserved.