Evidential query-by-committee active learning for pedestrian detection in high-density crowds

Abstract

Mascle. Evidential query-by-committee active learning for pedestrian detection in high-density crowds. International Journal of Approximate Reasoning, Elsevier, In press, AbstractThe automatic detection of pedestrians in dense crowds has become recently a very active topic of research due to the implications for public safety, and also due to the increased frequency of large scale social events. The detection task is complicated by multiple factors such as strong occlusions, high homogeneity, small target size, etc., and different types of detectors are able to provide complementary interpretations of the input data, with varying individual levels of performance. Our first contribution consists in outlining a fusion strategy under the form of an ensemble method, which models the imprecision arising from each of the detectors, both in the calibration and in the spatial domains in an evidential framework. Then, we propose a sample selection for augmenting the training set used jointly by the committee of classifiers, based on evidential disagreement measures among the base members in a Query-by-Committee context. The results show that the proposed fusion algorithm is effective in exploiting the strengths of the individual classifiers, as well as in augmenting the training set with informative samples which allow the resulting detector to enhance its performance.

Evaluating Crowd Density Estimators Via Their Uncertainty Bounds

“…We validated our proposed approach on high-density crowd images acquired at Makkah during Hajj [20]. Besides evaluating the proposed FE+LFE network, we compared it to U-Net [21], originally introduced for medical image segmentation and very effective even on relatively small training datasets as in our case (35 crowd images).…”

“…(1), in order to obtain better prediction accuracy (at the expense of To stress the independence of the proposed evaluation approach with respect to the classifier used, Fig. 1d shows the results of the density estimation obtained with SVM using active learning (AL) as in [20], where an SVM-ensemble is built iteratively by training SVMs with different descriptors on selected informative samples. The imprecision derives both from possible errors in the calibration procedure to obtain probability estimates out of SVM scores, and from the score heterogeneity in the image space.…”

A new soft likelihood function based on power ordered weighted average operator

“…Many methods are presented to update the status with collected information, including Beyasian updating rule, Dempster combination rule, negation, dynamic model such as DEMATEL, neural model, and so on . The likelihood function is first defined as the product of probabilities.…”

An evidential evaluation of nuclear safeguards

“…28 In the present work, a new method for evaluating nuclear safeguards is proposed. The method is based on Dempster-Shafer theory of evidence, which has been widely invoked to handle uncertain information, such as fault diagnosis, [29][30][31][32] human reliability analysis, 33 pattern recognition, [34][35][36][37] uncertainty modeling, [38][39][40] evidential reasoning, 41,42 decision making, 43 and environmental assessment. 44,45 In this article, first, according to the belief degrees of linguistic evaluation values given by IAEA experts and weights of IAEA experts about indicators, we obtain the basic probability assignment (BPA) of each indicator.…”