目的 在贝叶斯推理框架下，基于稀疏表示的跟踪算法能够较好地处理目标在视频场景中的各种复杂的外观变化，取得较为鲁棒的跟踪效果，但算法的计算复杂度很高，很难满足实时性要求。针对稀疏跟踪算法的这一问题，提出了一种基于L₂范数最小化的实时目标跟踪算法。方法 将主成分分析（PCA）子空间目标表示与L₂范数最小化进行结合，去除稀疏跟踪算法中常用的琐碎模板集，建立基于L₂范数最小化的目标表示模型以及将遮挡等因素考虑在内的观测似然度函数。结果 在大量的实验测试集上的对比实验结果显示，该算法和多个非常优秀的跟踪算法相比，可以达到相同甚至更高的跟踪精度，而且在多个测试集上可以达到20帧/s的速度。结论 该算法可以很好地应对视频监控场景中遮挡、光线突变、尺度变化和非刚性形变等干扰，同时算法复杂度低，满足了实时要求。

Objective Under the framework of the Bayesian inference,tracking methods based on sparse representations can deal with complex appearance changes in the video scene successfully and robustly.However,the computation costs are too expensive to achieve real-time tracking.To solve this problem,a new real-time tracking method based on L₂-norm minimization is proposed in this paper.Method The proposed method introduces the L₂ norm minimization into the PCA reconstruction,removes trivial templates from the sparse tracking method and presents an effective object representation model based on the L₂-norm minimization.An observation likelihood function that takes occlusion into account is designed in this paper.Result The experiments on many challenging image sequences demonstrate that the proposed method achieves the same and even better results when compared with several state-of-the-art tracking algorithms. Furthermore, it runs fast with a speed of about 20 frames/s.Conclusion The proposed method in this paper can handle occlusion,illumination changes,scale changes and no-rigid appearance changes effectively in video surveillance scenes with a lower computation complexity.Additionally, it can run in real-time.