Fast object tracking system is important to Augmented Reality (AR). However, tracking multiple objects fast and accurately is not a trivial task in computervision, especially when the objects become crowded and frequ...
详细信息
ISBN:
(数字)9781665453653
ISBN:
(纸本)9781665453653
Fast object tracking system is important to Augmented Reality (AR). However, tracking multiple objects fast and accurately is not a trivial task in computervision, especially when the objects become crowded and frequently occluded. In this paper, we design an efficient segmentation model along with the tracking and carefully study the contributions of pixel-level appearance modeling in the presence of occlusions. Different from previous joint tracking and segmentation methods that depend on bounding boxes for instance segmentation, we proposed an anchor free model to speed the object mask estimation. Specifically, our proposed approach jointly addresses instance segmentation and tracking with a single convolutional network for the first time, and generates object masks to explicitly separate overlapping objects in tracking. As pointed out by several previous works, the goals of tracking and segmentation are not always consistent, we deliberately design three homogeneous branches in our network to balance the competitive three functional parts of objectness, masks, and re-identification simultaneously and constrain them in a comparable scale. With the effective three functional parts, our method could run fast for the time-sensitive tracking task and meanwhile relieve the effects of appearance noise brought by the occlusions. Experimental results on the MOTS benchmark demonstrate the effectiveness of the proposed method.
Realistic augmented reality systems require both accurate localization of the user and a mapping of the environment. In a markerless environment this is often done with SLAM algorithms which, for localization, pick ou...
详细信息
ISBN:
(纸本)9781538675922
Realistic augmented reality systems require both accurate localization of the user and a mapping of the environment. In a markerless environment this is often done with SLAM algorithms which, for localization, pick out features in the environment and compare how they have changed from keyframe to current frame. However, human head agility, such as seen in video gaming tasks or training exercises, poses a problem;fast rotations will cause all previously tracked features to no longer be within the field of view and the system will struggle to localize accurately. In this paper we present an approach that is capable of tracking a human head's agile movements by using an array of RGB-D sensors and a reconstruction of this sensor data into 360 degrees of features that is fed into our SLAM algorithm. We run an experiment with pre-recorded agile movement scenarios that demonstrate the accuracy of our system. We also compare our approach against single sensor algorithms and show a significant improvement (up to 15 to 20 times better accuracy) in localization. The development of our sensor array and SLAM algorithm creates a novel approach to accurately localize extremely agile human head movements.
暂无评论