Joshua Kofi Asamoah

I am a PhD student at North Dakota State University, and a part of the Sustainable Mobility and Advanced Research in Transportation (SMART) Lab research group. Under the mentorship of Professor Armstrong Aboah, my research is focused on machine learning, deep learning, computer vision, and Internet of Things (IoT), particularly their applications in autonomous navigation and perception..

My current research project involves predicting lane intentions and vehicle trajectories using Naturalistic driving data. This work employs advanced computer vision and machine learning techniques to analyze real-world driving scenarios. Additionally, I'm exploring IoT integration to improve autonomous systems' perception, enabling safer and more efficient navigation in complex environments.

I have a BS in Civil Engineering from Kwame Nkrumah University of Science and Technology, where I worked in Dr. Jack Banahene's lab. Upon completion, I worked as a research and teaching assistant for Dr. Russell Afrifa.

Email  /  GitHub  /  Google Scholar  /  LinkedIn

News

[09/25/2024] Our research paper titled A novel methodological framework for assessing traffic sign retroreflectivity using Lidar data has been accepted for presentation at the TRB 2025 Conference

[09/21/2024] Our paper PaveCap: The First Multimodal Framework for Comprehensive Pavement Condition Assessment with Dense Captioning and PCI Estimation has been accepted for presentation at the TRB 2025 Annual Conference Meeting

[07/08/2024] Congratulations! The SMART Lab received a seed grant from AI SUSTEIN.

[07/02/2024] Congratulations! The SMART Lab was awarded an EDRF grant.

[06/10/2024] Excited to join the SMARTLab.

Research

computer vision, machine learning, transportation engineering

Intel Projects

Besides my work on the RealSense depth sensors and the publications above, a sampling of my publicly disclosed work

Intel RealSense 400

My responsibilities included system performance, components of the stereo algorithm on the imaging ASIC, and contributions to the design of laser projector pattern.

Compact VCSEL Projector

A low-cost dense, configurable projector system for RGB-D depth sensors.

Depth Image Enhancement

Algorithms to filter, enhance and clean-up RGB-D data streams.

Real-time Box Measurement

Using a single depth sensor, real-time detection of cuboids, accurate estimation of their dimensions, and even some bin-packing.

DashPoint: A low-cost, low-power human interface device

Finger tracking on a microcontroller, with optics tricks and some HCI ideas

Stereoscopic depth reconstruction with probabilistic pixel correspondence search

A fast method for performing stereo depth maps.

Other Projects

These include coursework, side projects and unpublished research work.

Dice Stacking: A Dynamic Manipulation Task

With Hunter Goforth, we designed a manipulation task and solved it with imitation learning.

Introspective Neural Networks

Using pre-trained neural networks to improve fine grained recognition via style transfer.

Stochastic Sampling of Parametric Policies

Using a very simple algorithm to solve some very simple environments

Optimizing for Physical Simulation

With Chris Atkeson and Alex Spitzer. Using optimizers to match an observed trajectory.

A Maze Bot

Making a 6-DoF PUMA arm solve a maze with real-time vision and tracking.

Learning Implicit Communication Strategies

Work with Aaron Goodman on used reinforcement learning to discover implicit collusion strategies in the context of an iterated prisoner’s dilemma.

Computational models for text summarization

Work with Ludwig Schubert on simplified encoders stages for text summarization.

Superresolution Micrscopy

An implementation of Faster STORM using compressed sensing.

Automatically building Restaurant Ontologies

Using the Yelp dataset of reviews to model the semantics and relationships between cuisines, businesses and other properties useful for restaurant recommendations.

Beyond Correlation Networks for the Financial Market

Using graph models, we track the development of financial networks over the 20th century.

Gradient-learned Models for Stereo Matching

Some re-implementations of standard stereo correspondence algorithms, along with experiments using classification for stereo matching.

Multimodal Natural Language Inference

We explored how natural language inference tasks can be augmented with visual data.

CNNs for 3D Model Classification

3D shape classification by learning an embedding function into a 2D image and using a pre-trained ImageNet network. At the time, got state-of-the-art results for single-view classification on ShapeNet40.

Wide-angle Stereo Lenses

We introduce various projection functions in the analysis of stereoscopic depth sensors.

Doctor Bayes

Detecting disease from a short description of symptoms. In some small testing, obtained nearly 90% top 5 accuracy and about 60% top 1 accuracy

Level-set based tracking and segmentation

We implemented a detection and deformable tracking pipeline for red blood cells.

Dequantization of Depth Data

An O(1) time algorithm for producing smooth normals for quantized data, such as the Kinect.

Golf swing monitoring

Work with Ankur Mehta, built a demonstration platform that used wireless low-weight, low-cost sensor platforms to monitor a golf swing.

Project Tetra: Collaborative robot state estimation

With Humphrey Hu, Ryan Julian, and Eric Yuan, a project to show the efficacy of multiple-robot collaborative state estimation. Using Wiimote cameras, mobile robot platforms, and real-time wireless communication.

GINA: Low power design

For testing and validating the functionality of the GINA (Guidance and Inertial NAvigation) mote, a 1.6 gram sensor platform.

GINA: Wireless sensor platform

I helped Anita Flynn and Thomas Watteyne build these small sensors and wrote firmware.

Design and source code from Jon Barron's website