Multi-stage Sampling with Boosting Cascades for Pedestrian Detection in Images and Videos

Abstract

Abstract. Many works address the problem of object detection by means of machine learning with boosted classifiers. They exploit sliding window search, spanning the whole image: the patches, at all possible positions and sizes, are sent to the classifier. Several methods have been proposed to speed up the search (adding complementary features or using specialized hardware). In this paper we propose a statisticalbased search approach for object detection which uses a Monte Carlo sampling approach for estimating the likelihood density function with Gaussian kernels. The estimation relies on a multi-stage strategy where the proposal distribution is progressively refined by taking into account the feedback of the classifier (i.e. its response). For videos, this approach is plugged in a Bayesian-recursive framework which exploits the temporal coherency of the pedestrians. Several tests on both still images and videos on common datasets are provided in order to demonstrate the relevant speedup and the increased localization accuracy with respect to sliding window strategy using a pedestrian classifier based on covariance descriptors and a cascade of Logitboost classifiers.

A coarse-to-fine approach for fast deformable object detection

“…By interpreting the score Sðx; yÞ as a posterior probability pðyjxÞ on the interpretations, inference can be reduced to the problem of drawing samples y from pðyjxÞ (because the most likely interpretations are also the ones with larger scores). Sampling ideas have been explored in the context of sliding-window object detectors in [14] demonstrating a two fold speed-ups over exhaustive search. Similar in spirit, but based on prior knowledge about the general shape of an object, are selective search [15] and objectness [16].…”

Crosstalk Cascades for Frame-Rate Pedestrian Detection

“…In the terminology of [15], every detector has a region of support (ROS) which is the neighborhood around a positive location in which the response remains positive. A detector's ROS is determined by the features, discriminability of the classifier, and alignment of the training data.…”

“…In Figure 3 (top) we show the ROS for the detector for various number of weak classifiers k. For every k the ROS has a non-negligible extent (σ ≥ 3). In previous work [14][15][16] a similar observation for the complete detector was used as a basis for fast detection schemes that compute a sparse set of responses and then sample more densely around promising locations.…”

“…Research on fast feature extraction [7,13,4] is complementary to our own. [14][15][16] propose to first compute a sparse set of detector responses and then sample more densely around promising locations; we use similar intuition for excitation of promising neighbors, but do so during every stage of detection for greater gains. Finally, research on optimizing other types of classifiers [1][2][3][17][18][19] is of considerable interest, however, such methods have difficulty matching the speed achieved by boosted classifiers.…”

Fast Pedestrian Detection by Cascaded Random Forest with Dominant Orientation Templates

“…It has been very successful in semantic image segmentation, key-point tracking, and object categorisation problems as a fast discriminative codebook. However, traditional classifiers like SVM [12,30] and Boosting [9,15] are still popularly adopted in the human detection domain, where RF is yet under-explored except for a few studies. Among them Hough Forest(HF) is a successful case [13,14].…”