Visual Object Tracking For Intelligent Human-computer Interaction

Locating and recognizing the object in video sequence,visual object tracking methods provide technically important foundation for applications of intelligent human-computer interaction.However,it is always difficult to guarantee the stability and efficiency of visual object tracking in complex and volatile environment.To address this issue,we did researches in aspect of basic algorithm,algorithm optimization and algorithm application,providing a complete reference solution to intelligent human-computer interaction application based on visual object tracking.Basic Algorithm: We propose a collaborative strategy that makes mutual enhancement between a discriminative tracker and a part-based tracker possible to obtain better overall performance.On one hand,we use validated results from the part-based tracker to update the discriminative tracker for recall performance improvement.On the other hand,based on confident results from the discriminative tracker we adaptively update the part-based tracker for simultaneous precision performance improvement.Experiments on various challenge sequences show that our approach achieved the state-of-the-art performance,which demonstrated the effectiveness of mutual collaboration between the two trackers.Algorithm Optimization:(1)Due to the negative effects on visual object tracking about which constantly changing background brings,we focused on extraction of discriminative feature and fast segmentation of target and applied background removal strategy based on feature matching to visual object tracking for performance improvement.And experimental results demonstrated that our method achieved better performance.(2)Based on the feasibility of applying background removal strategy to visual object tracking for performance improvement,we further investigated to improve the performance of collaborative tracker,using depth information obtained for a RGB-D camera to more efficiently remove the background.Moreover,based on the depth information and the object position returned by the collaborative tracker,we estimated the relative pose from the object to camera.Also,we applied the collaborative tracker based on RGB-D camera to upper body tracking and optimized the tracking procedure.Experimental results and analyses demonstrated that the upper body tracking method could provide a basic technique for applications of intelligent human-computer interaction.Algorithm Application: Based on PID controlling framework and the relative pose obtained by the collaborative tracker in upper body tracking,we implemented a following robot system FOLO with mobile robot platform ULBrain and development chassis Xbot by QFeel Tech independent research and development.We carried on tests on FOLO in an office environment and promising performances were obtained,which proves that our tracking method is applicable.