Taken during IPIN2015 in Banff, Canada

Joseph Menke

Taken during IPIN2015 in Banff, Canada

About me

I am a 2nd year PhD student in Electrical Engineering and Computer Science at the University of California, Berkeley. I work in the Video and Image Processing Lab, advised by Prof. Avideh Zakhor. My primary research interest lies in developing advanced localization algorithms for use in robotics and augmented reality

I graduated summa cum laude with a Bachelors of Science in Electrical Engineering from the University of California, Riverside. There I worked for the College of Engineering - Center for Environmental Research and Technology under Prof. Matthew Barth. I conducted research on identifying vehicles and other features in point cloud data and, as my senior design project, developed a fully autonomous electric utility truck along with two other students.

My LinkedIn page can be found here.

Current projects

High Accuracy Visual-Inertial Tracking for Smartphones

My current research is the development of a Smartphone based tracking algorithm that uses the phone's build in camera and inertial measurement unit to reconstruct the path which the device has traveled. These paths can be used for collecting data about the interiors of buildings that is otherwise very costly to collect, both in terms of time and equipment

Gaussian Mixture Models for Visual-Inertial Tracking

I am interested in developing efficient ways in which to apply Gaussian Mixture Models to Visual-Inertial Tracking. These models have the potential to reduce errors introduced by nonlinearities and non-gaussian noise by modeling the probability distribution over state variables as a Mixture of Gaussians rather than as a single gaussian (Kalman Filtering) or a set of samples (Particle Filtering). Traditional methods of training and utilizing Gaussian Mixture Models however are far too computationally complex to use in estimating the large number of variables needed for Visual-Inetial Tracking. Therefore new algorithms have to be developed for calculating the state distributions from visual and inertial information.

Previous projects

Multi-Modal Indoor Device Positioning

Using WiFi, Image, and Magnetic measurements we are able to localize and track a smartphone in a large building such as a mall or conference center with a mean accuracy of 2.5m. To do this we first build a database that consists of fingerprints of each type of measurment. This database is collected with a single walkthrough of the building at normal walking pace. The maps have shown to remain accurate for months after data collection, though the ease of database generation makes it possible to rebuild maps whenever necessary.

Undergraduate Senior Design: Autonomous GEM Electric Truck

Modified a GEM electric utility truck to autonomously follow GPS waypoints and avoid obstacles. The project was carried out over a period of six months start to finish by a team consisting of two other undergraduate students and myself. For this we designed a PCB for connecting an ARM microcontroller to the various control interfaces of the truck. We used two SICK LIDAR modules for detecting obstacles. A Trimble GPS module used RTK GPS to provide position estimates accurate within 5cm. We added a computer running realtime linux for handling path following and interfacing with the LIDAR and GPS modules. A blog of our work can be found here.