DADA: Driver Attention Prediction in Driving Accident Scenarios

Driver attention prediction is becoming an essential research problem in human-like driving systems. This work makes attempt to predict the d river xmlns:xlink="http://www.w3.org/1999/xlink">a ttention riving ccident scenarios (DADA). However, challenges tread on heels of that because dynamic traffic scene, intricate and imbalanced accident categories. In this work, we design a semantic context induced attentive fusion network (SCAFNet). We first segment RGB video frames into images with different regions (i.e., images), where each region denotes one category scene (e.g., road, trees, etc.), learn spatio-temporal features two parallel paths simultaneously. Then, learned are fused by find semantic-induced variation driver prediction. The contributions three folds. 1) With images, introduce their verify manifest promotion effect for helping prediction, modeled graph convolution (GCN) images; 2) fuse strategy, details transferred over convolutional LSTM module obtain map frame consideration historical situations; 3) superiority proposed method evaluated our previously collected dataset (named as DADA-2000) other challenging datasets state-of-the-art methods.