MODEL-BASED ROBUST AND PRECISE TRACKING EMBEDDED IN SMART CAMERAS—THE PFAAM-CAM

Top down tracking approaches like particle filtering are known for their robustness since they can handle multimodal proba-bility density functions. Active appearance models (AAMs), on the other hand, allow for precise, model-based tracking but suffer from limited robustness. The particle filter AAM combination (PFAAM), exploits the best of both worlds. In this paper the PFAAM is embedded on a smart camera. We present the necessary changes to achieve faster performance for the limited resources available in this embedded environ-ment. The proposed implementation of the PFAAM on the smart camera—the PFAAM cam—offers various benefits compared to a more traditional, centralised approach. All the process-ing is performed on the camera that allows it to run on the raw and thus artefact free video data. Also, only the resulting pa-rameters of the AAM are transmitted; no video feed has thus to leave the camera that significantly reduces the necessary bandwidth. Index Terms — tracking, particle filters, active appear-ance models, smart cameras 1