Current Projects Up for Bid

Robot arm and object localization using Deep Learning Segmentation Models

Recommended Discipline: Computer Science / Computer Engineering

Title: “Object localization for robot arm and item of interest using stereovision, lidar, and camera.”

Goal of Project: Provide a python API that will give x,y,z coordinates of robot arm and item of interest using a stereo camera and lidar.”

Description: A computer vision model will be needed to detect the robot arm and an item of interest.  Code will then have to be made to relate pixels from the camera to the lidar and/or stereovision point clouds to infer distance to both the arm and the item.  The data provided will be unlabeled so that it will have to be labeled and fed into a pretrained model for transfer learning.  After building a good vision model these points will have to be overlayed with the 3d points to provide coordinates.  Programming language will be Python.  Deep Learning framework will be Pytorch.  Deliverable will be an API that uses a Deep Learning model to provide x,y,z points for robot arm and item given lidar and or sterocamera as input data.

QGIS Plugin for Window-based Processing Library

Recommended Discipline: Computer Science

We have developed an open-source library for fast processing of high-resolution geospatial raster data. Conventional processing techniques that use 3×3 windows work well for traditional remotely sensed data such as Landsat with its 30m resolution. However, for medium and high-resolution data, such small window sizes would require an additional resampling step that reduces the resolution of the output data.  

Our techniques allow processing large windows in a time that is logarithmic in the window size. Our techniques can be used to derive topographic quantities like slope or curvature, statistical measures such as regression coefficients between bands, and even results of complex analysis, such as fractal dimensions. As such, we can process high-resolution data over large windows that contain relevant neighborhood information regardless of how many pixels must be doing so without losing output resolution.

The functionality is currently available in an open-source library that uses NumPy for vector-based processing of imagery. The goal of the proposed project is to create a QGIS plugin on its basis. Accomplishing this would give QGIS the capability of processing the medium-to-high resolution data from micro satellites and aerial remote sensing efforts without a need for resampling.

Current Algorithms
Our Library

Object Recognition on Traffic Data

Recommended Discipline: Computer Science

We are looking for building up a Machine learning Lightweight Web-application (Desktop and Mobile friendly) that uses a pre-trained TensorFlow.js  object detection API on traffic data.

Phase-1: Application should be able to detect the vehicle counts and type of vehicles from live traffic using mobile camera positioned at any height or angle.

Application should be able to detect the vehicle counts and type of vehicles when a recorded traffic video feed into the web application running on the desktop.

To train this model, we are looking for user friendly UI design for staff or students to feed data by uploading videos or images and to label them.

Expected Output: Van Class 2, Truck class 9, Car Class 2

North Dakota State University students working for Ericsson, Linkoping, Sweden (via Linkoping University)

Ericsson is a global leader in delivering Information and Communication Technology (ICT) solutions. In fact, forty percent of the world’s mobile traffic is carried over Ericsson networks. This semester’s project for Ericsson was to develop a mobile application that can send an android ping to a database. Once that ping is in the server, the web interface that we created will communicate with the web server we setup to display real-time data circles that entail cellular network information.