Multiple cameras pedestrian detection using feature fusion based on HIK SVM and removing phantoms with multi-view Bayesian network

This work presents a robust multiple cameras pedestrian detection approach with the fusion of integral channel features to improve the effectiveness and efficiency of pedestrian detection. The proposed method combines the histogram of oriented gradient (HOG) and local binary pattern (LBP) features by a concatenated fusion method. Appreciating with the low time complexity, we choose support vector machine (SVM) with the histogram intersection kernel (HIK) as a classifier which provides a good detection time while performing a suitable accuracy. Due to the heavy occlusions in real surveillancescenario, calibration errors and the diverse heights of pedestrians, phantoms (i.e., fake pedestrians) may be produced which degrades the detection performance. This paper utilizes multi-view Bayesian network model (MvBN) to remove phantoms. Given the preliminary results obtained by any multi-view pedestrian detection method, which are actually comprised of both real pedestrians and phantoms, the MvBN is used to model both the occlusion relationship and the homography correspondence between them in all camera views. As such, the removal of phantoms can be formulated as an MvBNinference problem. Moreover, to reduce the influence of the calibration errors and keep robust to the diverse heights of pedestrians, Height-adaptive projection (HAP) method is implemented to further improve the detection performance byutilizing a local search process in a small neighborhood of heights and locations of the detected pedestrians. Experimental results on four public benchmarks show that the proposed method out performs several state-of- the-art algorithmsremarkably and demonstrates high robustness in different surveillance scenes.

کلیدواژه ها:

Pedestrian detection ، histogram intersection kernel (HIK) ، multi-view Bayesian network (MvBN) ، Height-adaptive projection (HAP)