Jorge Martinez Castaneda

About

"I want to contribute in making our faculty a welcoming place to study, advance science, and work"

I'm an Assistant Professor with focus on Education at the Multimedia Computing Group of the Intelligent Systems department of EEMCS where I do research on improving speech recognition in adverse conditions. I received my PhD degree from TuDelft in 2013 and worked afterwards with research groups from Google and Huawei, as well as in other small companies in the Netherlands. I specialize in acoustical signal processing. In particular modeling and simulation of "multi-channel room impulse responses", which is a characterization of how the sound field evolves in a given place. This has many uses in our modern society, from improved speech recognition in voice- controlled "intelligent personal assistants", to better noise reduction in your mobile phone, or to be able to understand the playing message at the train station even when the train arrives. I currently maintain active collaboration with industrial partners Bang & Olufsen, Alpine, and Kien on related research topics.

I also have a heart for education. I’m motivated into pursuing an education vision that could be described as “emphatic and inclusive scientific education”. A vision that incorporates ways to getting to know and understand better our societal context and ourselves, our students and each other, to more effectively give students the long-lasting skills and values they need to cope with the challenges of modern society. I worked in the past as an educator in our faculty where I could gather insights into the current opportunities and challenges of our educational methods and programs.

Apart from being an educator and a scientist I consider myself an idealist and feminist, dreaming and striving at making education and science accessible and inclusive for everyone. I'm member of the faculty's Diversity and Inclusion Team (EDIT) and I'm member of the faculty's Work's Council (OdC).

Want to talk about any of these subjects? Do not hesitate in contacting me!

Publications

Courses

Courses 2024

Courses 2023

Courses 2022

Courses 2021

Courses 2020

Courses 2019

Courses 2018

Courses 2017

Courses 2015

Courses 2014

Courses 2013

News

Thesis Projects

Open Proposals

1. Interpreting the sound produced by cows using a potted microphone at the ear

Project Information

Starting: TBD
Industrial Partner: CowManager
Apply via E-mail

Problem Statement

In 2004, CowManager invented and launched the very first ear sensor for cows. Using accelerometer measurements at the cows ear, the activity or behaviour of the cow is determined. The behaviour information of a cow is used to determine whether or not a cow is healthy, pinpoint the moment a cow should be inseminated and analyse the feed intake of the cows.

Objectives

  1. The project can be divided into a technical and subjective part: one being the processing of noisy audio measurements from a potted microphone, the other being the interpretation of the obtained audio.

  2. The research must bring these two aspects together while taking into consideration the limited battery power available.

  3. To achieve this, you will create your own data set with custom hardware clipped to the cows being held at the farm in Harmelen.

Expected Outcomes

Automatically interpreting the sound produced by cows will give new insights in the well-being of a cow, especially its emotional state. This information can subsequently be used to inform farmers about unhappy or unwell cows among the herd.

2. Contact-less palm recognition for Bio-Metric login credential system

Project Information

Starting: November 2024
Industrial Partner: EYA Solutions
Apply via E-mail

Problem Statement

This thesis focuses on researching algorithms for video-based contact-less hand-palm recognition, across various lighting conditions with the aim to integrated it into a system that employs biometric authentication methods for auto-login (zero-interaction systems). This system is being developed for Elderly healthcare.

Objectives

  1. Conduct a literature research on state-of-the-art algorithms and applications regarding contact-less biometric data analysis (hand-palm detection) using video live-streams.

  2. Choose on possible algorithms candidates that are suitable for real-time embedded systems running on Single Board Computers (SBC).

  3. Research on improving the state-of-the-art algorithms to make them applicable in at different distances, different environmental light conditions, and crucially, suitable for parallelization, e.g., multiprocessing, multi-threading.
    Additionally, you will show the applicability of your algorithm in a real setup by integrating it into an existing system, and (co)create the necessary modules to achieve this.

Expected Outcomes
The system should be deployable as a subsystem using a shell script and must be isolated by an API from the rest of the system, with which it integrates this as an module into an existing backend and frontend. The algorithm should have a high accuracy and almost real-time performance.

3. Emotion Detection During a Video Chat for Healthcare Systems

Project Information

Starting: November 2024
Collaboration: EYA Solutions
Apply via E-mail

Problem Statement

This thesis involves researching algorithms analyzing and detection emotion based on facial expressions of elderly users during video chats. The aim is to seamlessly integrate this algorithm into an existing system, tracking their emotional state and interaction experience

Objectives

  1. Conduct a literature research on state-of-the-art algorithms and applications regarding contact-less biometric data analysis (hand-palm detection) using video live-streams.

  2. Choose on possible algorithms candidates that are suitable for real-time embedded systems running on Single Board Computers (SBC).

  3. Research on improving the state-of-the-art algorithms to make them applicable in at different distances, different environmental light conditions, and crucially, suitable for parallelization, e.g., multiprocessing, multi-threading.
    Additionally, you will show the applicability of your algorithm in a real setup by integrating it into an existing system, and (co)create the necessary modules to achieve this.

Expected Outcomes
The system should be deployable as a subsystem using a shell script and must be isolated by an API from the rest of the system, with which it integrates this as an module into an existing backend and frontend. The algorithm should have a high accuracy and almost real-time performance.

4. Multi-Modal BioMetric Data Fusion Such as Facial Recognition, Hand-Palm, and Fingerprints in Various Lighting Conditions for Auto Login Application Systems

Project Information

Starting: November 2024
Collaboration: EYA Solutions
Apply via E-mail

Problem Statement

This thesis involves researching the integration of biometric data from heterogeneous sensors for auto-login of the elderly into the system. It involves employing biometric authentication methods, including facial or hand-palm recognition including fingerprint recognition, across various lighting conditions.

Objectives

  1. Conduct a literature research on state-of-the-art algorithms and applications regarding contact-less biometric data and on efficient data fusion techniques. The goal is to best fuse the different biometric data modalities to realize seamless auto-login applications using these different sensor data.

  2. Choose on possible algorithms candidates that are suitable for real-time embedded systems running on Single Board Computers (SBC).

  3. Research on improving the state-of-the-art algorithms to make them applicable in at different distances, different environmental light conditions, and crucially, suitable for parallelization, e.g., multiprocessing, multi-threading.
    Additionally, you will show the applicability of your algorithm in a real setup by integrating it into an existing system, and (co)create the necessary modules to achieve this.

Expected Outcomes
The system should be deployable as a subsystem using a shell script and must be isolated by an API from the rest of the system, with which it integrates this as an module into an existing backend and frontend. The algorithm should have a high accuracy and almost real-time performance.

Completed Proposals

Projects as Supervisor

1. Visible Light Positioning for Balloon-Enhanced Drones

Author: Z. Lou (Electrical Engineering, Mathematics and Computer Science)

This thesis explores the integration of VLP with a balloon-enabled drone—a novel UAV setup featuring a buoyant balloon that extends flight duration. A balloon-enabled drone introduces both opportunities and challenges for VLP methods due to its size. Its large surface area can block light paths, which may impact signal reception and positioning accuracy. On the other hand, it also allows for the use of multiple receivers across the surface, potentially improving positioning reliability.

Read More

2. Using our tools backwards, AF detection by confusing time and frequency

Author: M. Kraaijeveld (Electrical Engineering, Mathematics and Computer Science)

This work explores the possibilities of reinterpreting speech processing techniques for use in atrial fibrillation detection. An existing method of modelling single heartbeat, single lead ECG signals by means of an ARMA model's amplitude response as a time domain signal is implemented. The parameters of the models are then used for AF detection by means of detecting P wave absence. For this detection, the distribution of the P wave associated parameters is compared to a GMM model of normal sinus rhythm beats obtained from a large number of recordings from different sources.

Read More

3. Feature extraction and classification on heart rate time series for cardiovascular diseases

Author: M. Beekhuizen (Electrical Engineering, Mathematics and Computer Science)

The study investigated if long-term wearable data can be used for the detection of heart diseases. The BigIdeasLab_STEP dataset and long-term Fitbit data from the ME-TIME study were used to examine this. Our analysis showed a correlation between the window/stride size and accuracy when performing activity classification with the BigIdeasLAB_STEP dataset. Moreover, variability was found between subjects due to differences in the physical structure of their hearts. Normalization proved to be a crucial step to minimize the subject variability and significantly improved performance. Grouping subjects and performing classification inside a group helped to improve performance and decrease inter-subject variability. Integrating handcrafted features in deep learning networks also improved classification performance.

Read More
Discover more from Jorge

Projects as Examiner

Share this page: