Driving posture recognition by joint application of motion history image and pyramid histogram of oriented gradients

This paper presents a novel approach to vision-based driving posture recognition. A driving action dataset was prepared by a side-mounted camera looking at a driver's left profile. The driving actions, including operating the shift lever, talking on a cell phone, eating and smoking, are decomposed into a number of predefined action primitives, which include operation of the shift lever, interaction with the driver's head and interaction with the dashboard. A global grid-based representation for the action primitives was emphasized, which first generate the silhouette shape from the motion history image, followed by application of the Pyramid Histogram of Oriented Gradients (PHOG) for more discriminating characterization. The random forest (RF) classifier was then exploited to classify the action primitives. Comparisons with some other commonly applied classifiers, such as kNN, multiple layer perceptron (MLP) and support vector machine (SVM), were provided. Classification accuracy is over 95% for the RF classifier in holdout experiment on the four manually decomposed driving actions.