Research And Implementation Of Human-computer Interaction Based On Visual Tracking Technology

With the development of computer vision technology, visual widely applied to many fields. Human-computer interaction based on visual tracking is a research hotspot in artificial intelligence and robotics in recent years. There are a lot of new methods appeared in the field of visual tracking, but many of these methods can only deal with simple context of the target tracking interaction. Delays, missed the targets, unsatisfactory tracking results also seriously affected the process of human-computer interaction. Implementation of visual tracking in the context of complex and real-time is a trend, which played an important role on human-computer interaction. Therefore, it is a great significance to research the human-computer interaction based on visual tracking.This paper researched key technologies and implement system in visual tracking algorithm enhance and platform. There are three improvements in segmentation algorithm design:firstly, on the basis of a comprehensive analysis of visual interactive features, using inter-frame difference method to get moving targets, realized natural visual interaction; secondly, view of the partition effect of the inter-frame difference method is not good enough, introduce the internal constraints of energy and local adaptive energy function to level set to solve the re-initialization of the problem, to improve the self-adaption; finally, for complex background image interference, fuzzy contours, target edge segmentation is unclear or over-segmentation problem, put forward a level set segmentation method which based on information integration, to overcome the traditional method of leakage segmentation and over-segmentation, obtain a good segmentation results. In the improvements of tracking algorithm, firstly, aim at the problem that LK flow method can not capture a larger movement of TLD (the Tracking Detecting Learning) algorithm, this paper used pyramid optical flow method, to get more corner information to overcome the aperture problem; secondly, for the random forest can not be randomly divided, this paper using extreme random forest method, introduce mixed samples to improve the classification accuracy. To build platform, based on the humanoid robot NAO, this paper build an interactive platform, tracking algorithm researched before is applied to the NAO robot, proposed tracking understand algorithm, identify interactive information to achieve human-computer interaction.Experimental results show that, improved segmentation algorithm is faster, better than before, improved tracking algorithm has high precision and real time, can be applied to human-computer interaction systems, the system has better robustness and higher accuracy.