Efficient Occlusion Handling Object Tracking System
C. Sathish, Merwin Amala Roger · 2014
Most object tracking algorithms use discriminative classifier which separates the target object from its background. This might lead to the inclusion of noisy samples when they are not sampled properly, causing tracking drift. This project work handles appearance change caused by intrinsic and extrinsic factors. A novel online object tracking algorithm is proposed which combines Principal Component Analysis (PCA) and L1 regularization. PCA is exploited with recent sparse representation schemes for learning effective appearance models. L1 regularization is introduced into the PCA reconstruction. The combination of this novel algorithm is developed to represent an object by sparse prototypes that accounts explicitly for data and noise. Occlusion and motion blur is taken into account in order to reduce tracking drift. It does not simply include the image observations for model update. Numerous experimental evaluations on challenging sequences demonstrate that the proposed tracking algorithm performs favorably well against several state-of-the-art algorithms.