Driver behaviour analysis using topological features

Driving behaviour prediction is a challenging problem due to the nonlinearity of human behaviour. Linear and nonlinear techniques have been used to solve this problem, and they provide good results presented in the performance of the current autonomous cars. However, they lack the ability to adapt to abruptness that happens because of the human factor. In this paper, we introduce a method to extract persistent homology barcode statistics. These statistics are useful as a representative of the driving process including the human behaviour. Human factor identification requires finding features that preserve certain properties against scalability, deformation, and abruptness. Topological Data Analysis (TDA) using persistent homology provides these features for driver behaviour prediction. We captured a driver’s head motion as an experimental behavioural cue, combined it with captured simulated vehicle data (location and velocities). Barcodes are extracted using JavaPlex, then we extracted descriptive statistics to show the significance of these barcode as features for driver behaviour prediction. The correlation between the extracted features shows a promising start for a behavioural tracking applications using TDA.