Hybrid Focal Stereo Networks for Pattern Analysis in Homogeneous Scenes

2019

Sensors

Self Cite

The main objective of this work is to study the applicability of ensemble methods in the context of deep learning with limited amounts of labeled data. We exploit an ensemble of neural networks derived using Monte Carlo dropout, along with an ensemble of SVM classifiers which owes its effectiveness to the hand-crafted features used as inputs and to an active learning procedure. In order to leverage each classifier’s respective strengths, we combine them in an evidential framework, which models specifically their imprecision and uncertainty. The application we consider in order to illustrate the interest of our Multiple Classifier System is pedestrian detection in high-density crowds, which is ideally suited for its difficulty, cost of labeling and intrinsic imprecision of annotation data. We show that the fusion resulting from the effective modeling of uncertainty allows for performance improvement, and at the same time, for a deeper interpretation of the result in terms of commitment of the decision.

Section: Resultsmentioning

confidence: 99%

Augmenting Deep Learning Performance in an Evidential Multiple Classifier System

2019

Sensors

Self Cite

“…Regression and validation sets I r , I v Init training set size M, batch size H Output: detectorL, mapping w, count error count P, hyperparameter w which could only settle for an inadequate compromise among the various sizes. Similarly to [12], we compute a perspective map M based on an accurate camera-to-ground pose estimation [1]. Then we are able to compensate the distortion by multiplying the detector score with the corresponding factor provided by the distortion map:…”

Section: Details Of Our Methodsmentioning

confidence: 99%

Active learning for high-density crowd count regression

2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS)

2017

Self Cite

Efficient crowd counting is an essential task in crowd monitoring, and significant advances have been made in this field recently by counting-by-regression techniques. We propose in this work a learning-to-count strategy with a generic detection algorithm which benefits from a counting regressor in order to identify crowded subregions with inadequate head detection performance, and to improve their representativeness in the training set. A straightforward but crucial step is proposed in order to take into account perspective correction within the proposed framework. An evaluation on Makkah images with medium to very high densities demonstrates the effectiveness of our algorithm and its capability to reach a count error of less than 5% in this difficult setting.

“…Query-by-bagging [27] or query-by-boosting [28] can be used to train weak classifiers on (weighted) randomly sampled variations of the training data set. Alternatively, a single model can be exploited and many variations of it can be 70 derived, e.g. changing its intrinsic parameters, like in [29] for naive Bayes, using the Dirichlet distribution over model parameters.…”

Section: Introductionmentioning

confidence: 99%

“…1.1. DatasetWe tested our proposed fusion method on high-density crowd images acquired at Makkah during Hajj[70]. The camera we used is a robotic camera (AVT Guppy PRO) mounted statically in order to observe the high-density pilgrim crowd, and providing gray-level regular images (visible spectrum).…”

mentioning

confidence: 99%

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

International Journal of Approximate Reasoning

2019

Self Cite

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.