Top–down visual attention integrated particle filter for robust object tracking

• A Top–down visual attention integrated particle filter (TAIPF) for tracking.
• A new top–down visual attention computational model based on frequency analysis.
• Deal with abrupt motion and longtime occlusion robustly.

Numerous tracking methods have been proposed and work well under many challenging conditions. However, there are still some problems need to be solved, such as abrupt motion and longtime occlusion. Visual attention mechanism enables humans to efficiently select the visual data of most potential interest and results in robust object tracking. Inspired by this fact, this paper presents a top–down visual attention computational model based on frequency analysis and integrates it into particle filter to solve the above mentioned problems. Given an image sequence, target-related salient regions are detected by the proposed top–down visual attention. Then the target is tracked by the proposed local and global search processes in which the salient regions are incorporated into particle filter. Comparison experiments on challenging sequences demonstrate the effectiveness of the proposed method.