Optimization Techniques for Automatic Assessment of Manually Drawn Diagrams in Educational Problem Sets

In STEM education, many problem sets require a student to answer a question with a drawing. For example, kinematic physics or statics/dynamics engineering problems often require the student to construct a free-body diagram as a necessary step in the solution. However, when such subjects are taught through online education (such as though a MOOC), most automatic assessment is only done through the constraints of multi-choice answers. The objective of this research project is to create an automatic assessment framework that can grade diagrams drawn and submitted by students through an online web interface. Our approach is a backend server-side module that decomposes the drawing into a set of shape primitives that are then automatically labelled and matched through an optimization approach to a set of solution primitives. Our approach must handle not only multiple correct solutions, but also select an appropriate grade based on a distance metric to the true solution.

Nathaniel Rossol
Faculty Supervisor: 
Osmar Zaiane
Partner University: