Predicting Driver Lane Change Maneuvers Using Driver's Face

DIPNet: Driver intention prediction for a safe takeover transition in autonomous vehicles

“…[2] and [20]) and multi-modal feature-fusion based approaches (refs. [6,13]- [18,21], and [22]). Overall, the data-driven approaches are efficient and costeffective when it comes to modeling large datasets with high prediction accuracy.…”

“…Refs. [6] and [18], are among the articles that only pay attention to the driver's observation video (in‐cabin video) and ignore the outside of the cabin. The models try to predict the driver's intention, using visual features such as head pose, eye movement, vehicles' dynamic information (e.g.…”

“…In ref. [18], Moussaid et al employed two processing sections to predict the intended maneuver. The first section focused on feature extraction using a CNN DenseNet121 [23] architecture, followed by obtaining a data frame with 256 attributes combined with the exterior features.…”

“…In ref. [18], in addition to head pose and eye movement features, the environment information is also fed to the model, manually. While the out‐cabin video can be extremely informative (like traffic lights status, repairs on the road, and accident situations on the road) and transmit information that the interior video does not [2].…”

“…Numerous studies have demonstrated that driver behavior , particularly eye movements, can be employed to assure safe takeover behavior in conditionally automated vehicles [10] as well as for activity recognition [11,12]. Other information such as head postures is also considered from video frames of driver observations in some literature [3,6,[13][14][15][16][17][18]. In some of these works, a mixture of information such as road traffic information [19] or car dynamics status is fed to the model as external data.…”