Introducing AI to Undergraduate Students via Computer Vision Projects

Kaiman Zeng, Yancheng Li, Yida Xu, Di Wu, Nansong WuAAAI 2018 · DOI

Abstract

Artificial Intelligence (AI) education often faces the challenge of being perceived as overly abstract by undergraduate students. This paper presents a pedagogical framework that introduces AI through a practical, mobile landmark recognition project. By engaging students in the full research lifecycle—from data collection to feature matching—the approach successfully improved technical proficiency and motivation, particularly among first-generation college students.

Introduction

For many undergraduates, AI can appear as an impenetrable "black box" of complex mathematics. This perception can be a significant barrier, especially for students from underrepresented backgrounds who may find initial academic hurdles discouraging. To address this, the authors developed a curriculum centered on a relatable, high-impact problem: navigating a college campus using mobile computer vision.

Methodology: Mobile Landmark Recognition

The project focused on the technical challenges of mobile landmark recognition. Unlike general object recognition, mobile systems must handle varying lighting, diverse viewing angles, and hardware constraints like latency and privacy. Students explored the trade-offs between server-side processing and on-device resource management.

Systematic Data Collection

A critical component of the project was the creation of a robust dataset. Students moved beyond random photography to design a systematic imaging plan. Each of the 23 targeted campus buildings was photographed from eight cardinal directions and three different distances.

Systematic imaging plan showing the eight cardinal directions used for data collection.

This structured approach ensured the dataset captured the diversity of real-world environments, including varying lighting conditions (morning vs. afternoon). The resulting Campus Landmarks Dataset consists of 3,725 images, each tagged with GPS coordinates.

Feature Extraction and Matching

The technical core of the project involved Speeded Up Robust Feature (SURF) extraction. Students learned how to detect "local salience"—unique points in an image that remain recognizable despite scaling, rotation, or lighting changes.

Visualization of SURF feature matching between query and reference images of campus buildings.

By performing experiments matching "query" images against the reference database, students gained an intuitive understanding of how algorithms distinguish between relevant landmarks and irrelevant background structures.

Key Insights

The project highlighted several important pedagogical and technical lessons:

  • Relatability as a Hook: Connecting AI to familiar technologies like augmented reality or social media tagging lowered the barrier to entry for complex mathematical concepts.
  • The Role of Context: Students discovered that mobile computer vision is enhanced by context. Integrating GPS and gyroscope data can significantly narrow the search space, improving both speed and accuracy.
  • Concrete Learning: Abstract concepts like "feature extraction" became tangible once students applied them to recognize their own campus library.

Educational Impact and Results

The initiative demonstrated significant educational benefits. Students showed marked improvement in cognitive and critical thinking skills, particularly regarding mathematical modeling. Furthermore, the hands-on research experience increased student interest in graduate studies and improved interpersonal communication. Technical validation confirmed that the student-collected dataset was sufficient for robust image matching using SURF features.

Limitations and Future Directions

While successful as a pedagogical tool, the current framework focuses on classical feature extraction rather than modern deep learning architectures like Convolutional Neural Networks (CNNs). Future iterations of the project aim to incorporate more complex simulations, such as self-driving car environments, and further integrate AI literacy across the undergraduate engineering curriculum.

Summary updated

BibTeX
@inproceedings{zeng2018introducing,
  author = {Kaiman Zeng and Yancheng Li and Yida Xu and Di Wu and Nansong Wu},
  title = {{Introducing AI to Undergraduate Students via Computer Vision Projects}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.1609/aaai.v32i1.11403},
}
Figure