Research On Visual Tracking Algorithms And Applications

Visual tracking is the core technology of city security,intelligent transportation and human-computer interaction.For the past few years,visual tracking enjoys a wide popularity with the development of high-performance computers,the usage of high quality cameras and the growing requirement of automated video analyses.The most challenging thing of visual tracking is that it has to handle various environments and challenging factors in real-time.This thesis focuses on developing efficient,accurate and robust tracking algorithms and applying stable and real-time visual tracking algo-rithm to a mobile robot platform.The main contributions of this thesis can be summa-rized as follows:1.A robust and efficient human tracking algorithm is proposed from the perspec-tive of.object representation.A hierarchical ensemble framework is construct-ed which can incorporate information including individual pixel features,local patches and holistic target models.This method can handle severe deformation as well as occlusion and perform excellently for moving human tracking with a speed in excess of 30 frames per second.2.An accurate and efficient object tracking algorithm is proposed from the per-spective of tracking model.This method absorbs the strong discriminative ability from structured output SVM and speeds up by the correlation filter algorithm sig-nificantly.A multimodal target detection technique and high-confidence model update strategy are proposed to improve the target localization precision,which prevent model drift and relieve the model corruption problem.The proposed tracker can handle the severe occlusion as well as object missing and performs superiorly against several state-of-the-art algorithms while runs at speed in excess of 80 frames per second.3.A stable and real-time 3D object tracking system is implemented on a mobile robot platform.The system fuses a monocular camera with an ultrasonic sen-sor to obtain the real-time 3D positions of a target.The monocular camera is employed to provide 2D partial location estimation.The ultrasonic sensor array is used to provide the range information.The extended Kalman filter is adopt-ed to sequentially process multiple,heterogeneous measurements arriving in an asynchronous order from the vision sensor and the ultrasonic sensor separately.