Robust Tracking with and Beyond Visible Spectrum: A Four-Layer Data Fusion Framework

Perceptual stimulus — A Bayesian-based integration of multi-visual-cue approach and its application

“…Results show that LogOP is more sensitive to the error than LOP, but it can increase the reliability of the estimation, which is depicted in Figure 4(b). According to this finding, we proposed an adaptive integrating strategy [6] which switches between the LOP and LogOP according to the reliabilities of cues. The proposed method has been applied successfully in hand tracking in a complex scene.…”

“…From the 1950s to the present, there appear two major branches in addressing the problem of inferring with uncertainties in the filed of Artificial Intelligence: D-S evidence theory and Bayesian method [6] . D-S evidence theory uses a belief function to determine the uncertainty of the information.…”

“…The authors proposed a four-layer fusion framework for robust tracking [6] . Sigal et al [39] proposed to track an articulated object with a high dimensionality by using a PGM to integrate information from each joint of the object; Beal et al [40] used a PGM to integrate the audio cue with video cue for positioning the speaker in the meeting room; Khan et al [41] used LogOP to integrate the color, motion and spatial cues for the scene segmentation; Wang et al [42] built a conditional random field in pixel-level to integrate the color cue with the motion cue to segment moving objects from their shadows.…”

“…We propose a particle filtering based adaptive IMVC method which choose the integrating strategies adaptively to the changes of the tracking scene [6] : We measure the uncertainty of the cue with its 2nd moment which can be calculated with the assigned particles; if the 2nd moment is too large, then the integrating strategy switches to LOP, otherwise it switches to LogOP.…”