Date of Award


Document Type

Honors Thesis (Open Access)


Colby College. Computer Science Dept.


Professor Oliver Layton


Humans have a remarkable ability to estimate their direction of self-motion, or heading, based on visual input stimulus (optic flow). Machines, on the other hand, have a difficult time with this task, especially when flow is introduced that is inconsistent with the motion of the observer. For example, when moving objects enter the field of view, their motion provides inconsistent flow data which often disrupts heading estimates of current heading estimation models. We investigate the ability of neural networks to estimate heading from optic flow data and the limitations of these models when different variations of inconsistent flow are introduced. Since convolutional neural networks (CNNs), in particular, have been shown to be ideal for image and video analysis tasks, we focus our exploration on neural networks with convolutional architectures. We show that a simple CNN can be trained to effectively estimate heading from optic flow image data with human-like accuracy. We further investigate the advantages of using a convolutional recurrent neural network (RNN) as opposed to a simple CNN to estimate heading. Our findings suggest that the ability of an RNN to take into account the temporal dimension provides an advantage when it comes to several types of inconsistent flow data. These advantages may justify the added complexity and more costly training time of an RNN versus a CNN when it comes to estimating heading from Optic Flow.


Neural Networks, Optic Flow, Autonomous Systems, Navigation