Date of Award
2025
Document Type
Honors Thesis (Open Access)
Department
Colby College. Computer Science Dept.
Advisor(s)
Oliver Layton
Abstract
Human driving is a complex visuomotor task and the specific visual clues that guide it remain under investigation. While prior research has emphasized gaze-based strategies such as the Tangent Point and Future Path hypotheses, recent evidence highlights the potential role of optic flow, the visual motion pattern perceived during self-movement, as critical to steering ability. This thesis explores whether raw optic flow alone can support accurate predictions of human steering behavior. We trained a convolutional neural network to map optic flow vector fields to steering angles in a virtual reality driving simulation. The dataset, collected by Giguere et al., included over 100,000 frames of the simulated road per subject and the subject's steering angle data. Our CNN performed well in short, high-density optic flow environments, capturing the shape and timing of human steering. However, performance deteriorated in low-density scenes and over longer driving sequences. The findings support optic flow’s relevance in steering behavior and lay groundwork for future models incorporating time integration and reaction delay alignment.
Keywords
neural networks, optic flow, artificial intelligence, deep learning, driving, cars
Recommended Citation
Bernard, Katie E., "Predicting Human Steering from Optic Flow Using Deep Convolutional Neural Networks" (2025). Honors Theses. Paper 1475.https://digitalcommons.colby.edu/honorstheses/1475