People and Object Tracking computer vision algorithms are used for tracking objects (eg. humans, vehicles, etc) in camera images for visual surveillance purposes. It can be widely used in sports video analysis, intelligent video surveillance, city traffic crossroads and pedestrian monitoring.
Person Tracking with Discriminative Appearance Modeling API
Our algorithm can handle several challenging cases in tracking application such as occlusion, pose changes, illumination, fast motion, cluster, distracter with similar appearance, etc. It works well over different resolutions and scenes.
1. Yuwei Wu, Mingtao Pei, Min Yang, Junsong Yuan, and Yunde Jia. Robust Discriminative Tracking via Landmark-based Label Propagation. IEEE Transactions on Image Processing (TIP), 2015
2. Min Yang, Yuwei Wu, Mingtao Pei, Bo Ma, and Yunde Jia. Online Discriminative Tracking with Active Example Selection. IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), 2015
We provide a powerful online discriminative tracking algorithm based on Laplacian Regularized Least Squares (LapRLS) . The tracking algorithm contains two steps. First, a manifold regularized semi-supervised learning method (i.e. LapRLS) is used to learn a robust classifier to detect the target object. Second, an active example selection approach is adopted to automatically select the most informative examples for LapRLS to ensure the high classification confidence of the classifier. The overview of the approach is shown below:
We evaluated this API with 11 state-of-the-art methods on a recent popular benchmark  where each tracker is tested on 51 challenging videos (more than 29,000 frames). The state-of-the-art trackers include the TLD tracker , tracking with Multiple Instance Learning (MIL) , Visual Tracking Decomposition (VTD) , the Struck method , the Sparsity-based Collaborative Model (SCM) , Laplacian Ranking Support Vector Tracking (LRSVT) , Compressive Tracking (CT) , Structural Part-based Tracking (SPT) , Least Soft-threshold Squares Tracking (LSST) , Randomized Ensemble Tracking (RET)  and tracking with Online Non-negative Dictionary Learning (ONNDL) . Comparison with the state-of-the-art trackers on the comprehensive benchmark demonstrates that our tracking algorithm is more effective and accurate.
We also provide a real time algorithm with high performance for fast person or object tracking requirements. This real time algorithm can achieve the speed of 0.031 second per frame. It adopts histogram to fit feature distribution and feature selection mechanism to further delete features which are less discriminative and improves the feature quality. Because of these two steps, the feature matching can be accelerated significantly and the tracking accuracy and robustness can be improved.
The following videos show the performance of our approach (bold, red box) against the other approaches mentioned above.