Visual Object Tracking
Our System is an advanced tracker that continuously tracks arbitrary objects in videos. Given the bounding box of any object in any video our artificially intelligent algorithm will track that object through the rest of the video. The tracker can adapt to changes in object appearance, size, lighting conditions etc. This tracker is the winner of VOT2018 public dataset. The key innovations of this tracker include adaptive spatial feature selection using a deep neural network trained offline and temporally consistent constraints, with which the new tracker enables joint spatio-temporal filter learning in a lower dimensional discriminative manifold. The process of learning spatial filters is formulated as lasso regularisation. To encourage temporal consistency, the filter model is restricted to small incremental changes and updated locally to preserve the global structure in the manifold.