Human detection using partial least squares analysis

By William Robson Schwartz, Aniruddha Kembhavi, David Harwood, Larry S. Davis. 2009

559 citations recorded by the source corpus.

Significant research has been devoted to detecting people in images and videos. In this paper we describe a human detection method that augments widely used edge-based features with texture and color information, providing us with a much richer descriptor set. This augmentation results in an extremely high-dimensional feature space (more than 170,000 dimensions). In such high-dimensional spaces, classical machine learning algorithms such as SVMs are nearly intractable with respect to training. Furthermore, the number of training samples is much smaller than the dimensionality of the feature space, by at least an order of magnitude. Finally, the extraction of features from a densely sampled grid structure leads to a high degree of multicollinearity. To circumvent these data characteristics, we employ Partial Least Squares (PLS) analysis, an efficient dimensionality reduction technique, one which preserves significant discriminative information, to project the data onto a much lower dimensional subspace (20 dimensions, reduced from the original 170,000). Our human detection system, employing PLS analysis over the enriched descriptor set, is shown to outperform state-of-the-art techniques on three varied datasets including the popular INRIA pedestrian dataset, the low-resolution gray-scale DaimlerChrysler pedestrian dataset, and the ETHZ pedestrian dataset consisting of full-length videos of crowded scenes.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Pedestrian Detection with Unsupervised Multi-stage Feature Learning (2012, methodology): “All existing state-of-the-art methods use a combination of hand-crafted features such as Integral Channel Features [9], HoG [5] and their variations [13, 33] and combinations [38], followed by a trainable classifier such as SVM [13, 28], boosted classifiers [9] or random forests [7].”
Deep roto-translation scattering for object classification (2014, methodology): “s a linear operator to reduce the number of scattering coefﬁcients and decorrelate them, before computing an SVM classiﬁer. The feature selection is implemented with a partial least square regression [22,31,24]. A logarithm non-linearity is applied to scattering coefﬁcients in order to separate low ”
Pedestrian Detection: An Evaluation of the State of the Art (2012, methodology): “Schwartz et al. [69] represented pedestrians by edges, texture, and color and applied partial least squares to project the features down to a lower dimensional space prior to SVM training [P LS ].”
Recent advances and trends in visual tracking: A review (2011, methodology): “Schwartz et al. [62] presented an efﬁcient descriptor for pedestrian detection based on Partial Least Squares (PLS) analysis.”
Max-Margin Object Detection (2015, methodology): “Many recent approaches include features for part-based-modeling, methods for combining local features, or dimensionality reduction [6, 20, 16, 4, 18].”
Fast Feature Pyramids for Object Detection (2014, unknown): “PLS [75] 40 62 71 55 57”
New features and insights for pedestrian detection (2010, background): “CSS Several authors have reported improvements by combining multiple types of low-level features [5, 18, 27].”
Optimized Pedestrian Detection for Multiple and Occluded People (2013, methodology): “As shown in Figure 1 using the PLS detector [20] with its default non maximum suppression method, correct candidate detections were present that were suppressed by stronger neighboring detections.”
A Survey of Multi-View Representation Learning (2016, unknown): “CCA can be interpreted as the maximum likelihood solution to a probabilistic latent variable model for two Gaussian random vectors.”
Discriminative Decorrelation for Clustering and Classification (2012, unknown): “In particular, Ke and Sukthankar [11] and Schwartz et al [12] examine (linear) low-dimensional projections of oriented gradient features .”
Competition on counter measures to 2-D facial spoofing attacks (2011, background): “Such a description can be obtained with the combination of feature channels focusing on different image characteristics, such as shape, color, and texture [24].”
Joint Deep Learning for Pedestrian Detection (2013, methodology): “The compared approaches are VJ [49], Shapelet [42], PoseInv [28], LatSVM-V1 [17], LatSVM-V2 [17], HikSVM [31], HOG [5], MultiFtr [52], HogLbp [51], Pls [43], MultiFtr+CCS, MultiFtr+Motion [50], FeatSynth [1] FPDW [10], ChnFtrs [11], MultiResC [37], CrossTalk [9], DN-HOG [34] and ConvNet-U-MS”
Color attributes for object detection (2012, unknown): “On this dataset, our approach yields a significant gain of 14% in mean AP over conventional state-of-the-art methods.”
Efficient Pose-Based Action Recognition (2014, methodology): “The weights w can be learned by partial least squares (PLS) [24], which has been recently adopted in computer vision for different applications including pose estimation and regression [25], image classification [26], pedestrian detection [27], and multiview learning [28].”
Multiresolution Models for Object Detection (2010, background): “Recent work has focused on models for handling pose variation [15, 11, 16–18], reducing complexity of learning [19, 20]), and multicue combination [21, 22].”
Robust Multi-resolution Pedestrian Detection in Traffic Scenes (2013, methodology): “In this part, we compare the proposed method with other state-of-the-art methods evaluated in [14], including: Viola-Jones [40], Shapelet [44], LatSVM-V1, LatSVM-V2 [19], PoseInv [30], HOGLbp [43], HikSVM [31], HOG[6], FtrMine [13], MultFtr [44], MultiFtr+CSS [44], Pls [37], MultiFtr+Motion [44], FP”
Human detection using depth information by Kinect (2011, unknown): “There has been much research in the past few years in human detection and various methods are proposed [1] [2] [6] [13].”