Feature Detection and Tracking with Constrained Local Models

摘要: We present an efficient and robust model matching method which uses a joint shape texture appearance to generate set of region template detectors. The is fitted unseen image in iterative manner by generating templates using the current parameter estimates, correlating with target response images optimising parameters so as maximise sum responses. similar that used Active Appearance Model due Cootes et al. However our approach likely feature templates, instead trying approximate pixels directly. show when applied human faces, constrained local (CLM) algorithm more accurate than original AAM search method, relies on reconstruction error update parameters. demonstrate improved localisation accuracy two publicly available face data sets tracking challenging in-car sequences.