Abstract
Accurate driver attention prediction can serve as a critical reference forintelligent vehicles in understanding traffic scenes and making informeddriving decisions. Though existing studies on driver attention predictionimproved performance by incorporating advanced saliency detection techniques,they overlooked the opportunity to achieve human-inspired prediction byanalyzing driving tasks from a cognitive science perspective. During driving,drivers' working memory and long-term memory play crucial roles in scenecomprehension and experience retrieval, respectively. Together, they formsituational awareness, facilitating drivers to quickly understand the currenttraffic situation and make optimal decisions based on past driving experiences.To explicitly integrate these two types of memory, this paper proposes anAdaptive Hybrid-Memory-Fusion (AHMF) driver attention prediction model toachieve more human-like predictions. Specifically, the model first encodesinformation about specific hazardous stimuli in the current scene to formworking memories. Then, it adaptively retrieves similar situational experiencesfrom the long-term memory for final prediction. Utilizing domain adaptationtechniques, the model performs parallel training across multiple datasets,thereby enriching the accumulated driving experience within the long-termmemory module. Compared to existing models, our model demonstrates significantimprovements across various metrics on multiple public datasets, proving theeffectiveness of integrating hybrid memories in driver attention prediction.