You can access the project by clicking on this link.

Overview

In an academic collaboration for the course LINFO2275 - Data Mining and Decision Making at UCLouvain, my colleagues Chloé Marchal, Mathieu Graf and I embarked on a challenging and insightful journey into the world of gesture recognition. Our project’s aim was to develop a smart user interface capable of recognizing hand gestures with high accuracy. We explored and implemented two distinct methodologies: the traditional edit-distance approach and the more contemporary convolutional neural networks (CNNs).

Project Highlights

  • Datasets: Engaged with two public datasets comprising 1000 sequences each, capturing ten subjects drawing ten different objects in 3D space.
  • Preprocessing: Mastered complex preprocessing techniques tailored to each method, converting raw data into usable formats for machine learning models.
  • Edit-Distance Method: Delved into the intricacies of dynamic programming to implement the edit-distance algorithm, enhancing it with k-nearest neighbors (knn) classification.
  • CNN Implementation: Designed and trained a convolutional neural network, adapting its architecture for the task of image-based gesture recognition.
  • Evaluation: Employed user-dependent and user-independent cross-validation strategies to assess model performance, gaining valuable insights into their applicability in real-world scenarios.

Learning Outcomes

  • Data Preprocessing: Acquired proficiency in transforming sequences into standardized formats and visual representations, laying a strong foundation for any machine learning workflow.
  • Algorithmic Understanding: Gained a deep understanding of the edit-distance algorithm and dynamic programming, recognizing their strengths and limitations in practical applications.
  • Cross-Validation Techniques: Enhanced my knowledge of model validation, learning how to reliably evaluate model performance and generalize findings beyond the training data.
  • Comparative Analysis: Learned to critically compare different models, not only in terms of their accuracy but also considering their suitability for different user scenarios.

Reflection

This project was a testament to the power of data mining and decision-making techniques in creating intelligent systems. Through iterative development and rigorous testing, I have become proficient in recognizing patterns within data, selecting appropriate machine learning models, and tuning them to meet specific performance criteria. The experience has solidified my understanding of the end-to-end process of developing a machine learning project, from conception to evaluation.

As a data science enthusiast, this project has been pivotal in sharpening my technical skills and deepening my appreciation for the field’s complexity and potential. It stands as a highlight in my portfolio, showcasing my ability to tackle advanced problems and derive actionable insights through data-driven methodologies.