Modeling the Effect of Driver’s Eye Gaze Pattern Under Workload: Gaussian Mixture Approach

AbstractThis paper puts forward a Gaussian Mixture Model (GMM) for eye gaze behavior under workload and applies it to the analysis of gaze distributions in an automotive context. Specifically, it extends earlier work on Information Constrained Control (ICC) by generating an ICC GMM derivative. We suggest a measure for workload estimation based on the Kullback Leibler divergence ($D_{kl}$) between tested eye gaze distributions and a reference workload-free distribution. This derivative assumes diagonal Gaussians that are distant from each other. Under these assumptions, we achieve an analytical measure that has significantly fewer parameters than discrete grid-like distributions. Testing our measure on eye gazing data collected during real world driving experiments in a highway environment confirms the effectiveness of this approach.

Return to previous page