Visual Attention-Based Object Detection in Cluttered Environments

The study of human visual attention is considered a hot topic in the field of activity recognition, experimental psychology research and human computer interaction. The importance of detecting user objects of interest in real time is critical to provide accurate cues about the user intentions. However, current methods for visual attention extraction and object detection suffer from low performance when moving to ongoing condition. Inherent complexity of cluttered environments is considered the major barrier to achieve good performances. To address this challenge, we present a novel method that includes head-worn eye tracker and egocentric video. Our method exploits sliding window-based time series approach in conjunction with a Heuristic probabilistic function to analyse user fixations around potential object of interest in an egocentric video. We evaluate the proposed method using a new dataset annotated with user gaze data and object within a frame image. Our experimental results show that our approach can outperforms several state-of-the-art commonality visual attention-based object detection methods.