Tracking Of Moving Object Based On Multiple Instance Learning

Visual tracking is a critical step in many computer vision applications such as surveillance, driver assistance systems and human-computer interactions. However, designing robust tracking method is still an open issue, which includes abrupt object motion, changing appearance patterns of the object and the scene, non-rigid object structures, object-to-object and object-to-scene occlusions. Multiple instance learning performs very well in object tracking. This paper proposes two novel methods based on multiple instance learning.Co-training Multiple Instance Learning (CoMIL) is based on the Co-training approach which labels incoming data continuously, and then uses the prediction from each classifier to enlarge the training set of the other. The discriminative classifier is implemented using online multiple instance learning (MIL), which can deal with inaccurate positive samples in the updating process and allow some flexibility while finding a decision boundary. Firstly, two classifiers are improved mutually in our CoMIL tracking system. Secondly, the update mechanism of CoMIL uses multiple potential positives according to the MIL which handles the update error due to the risk of extracting only one positive example. In addition, IVT as a generative model is an assistant tracker to form a simple cascade and can help the system avoiding wrong update. Experiments show that CoMIL tracking algorithm performs better than several state-of-the-art tracking algorithms on challenging sequences.Online Support Instance Tracking (OSIT) algorithm is in which the instances are selected adaptively within the multiple instance learning framework. The instances are selected by mapping instance-bag to train 1-norm support vector machines online. These selected support instances as a set locate the target in the next frame. An update algorithm is proposed to cope with pose variations and appearance changes. The update process includes retraining 1-norm SVM and reselecting support instances. In order to retain previous support instances information we introduce a forgetting factor. Experimental results demonstrate that OSIT is robust in handling occlusion, abrupt motion and illumination.