Samples Selective Updating Mechanism for Object Tracking

In the issue of object tracking, tracking method probably receives wrong position of an object due to the complex environment conditions in a real scenario, such as object occlusion, illumination change, and object deformation. For discriminative model tracking method, this wrong position information may results in tracking failure. To tackle the wrong tracking problems, we revisit the efficient convolution operators (ECO) method and introduce a samples selective updating mechanism. This updating mechanism employs cascaded screening method which can eliminate the wrong tracking output and protect the training set from being contaminated. The experiments are performed on OTB100 benchmark. And the test shows that our tracker achieves the OPE score of 93.5% and improves 1.19% compares with ECO.