Conference Paper

Detecting Pedestrians by Learning Shapelet Features

DOI: 10.1109/CVPR.2007.383134 Conference: 2007 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2007), 18-23 June 2007, Minneapolis, Minnesota, USA
Source: DBLP

ABSTRACT In this paper, we address the problem of detecting pedes- trians in still images. We introduce an algorithm for learn- ing shapelet features, a set of mid-level features. These fea- tures are focused on local regions of the image and are built from low-level gradient information that discriminates be- tween pedestrian and non-pedestrian classes. Using Ad- aBoost, these shapelet features are created as a combina- tion of oriented gradient responses. To train the final classi- fier, we use AdaBoost for a second time to select a subset of our learned shapelets. By first focusing locally on smaller feature sets, our algorithm attempts to harvest more use- ful information than by examining all the low-level features together. We present quantitative results demonstrating the effectiveness of our algorithm. In particular, we obtain an error rate 14 percentage points lower (at10−6 FPPW) than the previous state of the art detector of Dalal and Triggs (1) on the INRIA dataset.

57 Reads
  • Source
    • "In [17], they proposed improvement of edge-based feature (e.g. Haar- like[16], shapelets[18], and shape context[19]) by combining color self-similarity and the motion features from optic flow. However, this approach cannot be realized in real-time whose properties are obviously required for traffic safety systems. "
    [Show abstract] [Hide abstract]
    ABSTRACT: The paper presents a fine-grained walking activity recognition toward an inferring pedestrian intention which is an important topic to predict and avoid a pedestrian's dangerous activity. The fine-grained activity recognition is to distinguish different activities between subtle changes such as walking with different directions. We believe a change of pedestrian's activity is significant to grab a pedestrian intention. However, the task is challenging since a couple of reasons, namely (i) in-vehicle mounted camera is always moving (ii) a pedestrian area is too small to capture a motion and shape features (iii) change of pedestrian activity (e.g. walking straight into turning) has only small feature difference. To tackle these problems, we apply vision-based approach in order to classify pedestrian activities. The dense trajectories (DT) method is employed for high-level recognition to capture a detailed difference. Moreover, we additionally extract detection-based region-of-interest (ROI) for higher performance in fine-grained activity recognition. Here, we evaluated our proposed approach on " self-collected dataset " and " near-miss driving recorder (DR) dataset " by dividing several activities– crossing, walking straight, turning, standing and riding a bicycle. Our proposal achieved 93.7% on the self-collected NTSEL traffic dataset and 77.9% on the near-miss DR dataset.
    IEEE Intelligent Transportation Systems Conference (ITSC); 09/2015
    • "Object detection is usually a two-phase process, where binary classification is followed by feature extraction from a local 30 image patch. Different image features (Viola and Jones, 2004; Ahonen et al., 2006; Berg and Malik, 2001; Sabzmeydani and Mori, 2007; Cheng et al., 2006) and binary classifiers (Freund, 2001; Breiman, 2001) are available. To our knowledge such an image processing approach has not been applied to automate peak picking in which it 35 could provide a reliable algorithm for the identification of peaks in multidimensional NMR spectra. "
    [Show abstract] [Hide abstract]
    ABSTRACT: A detailed analysis of multidimensional NMR spectra of macromolecules requires the identification of individual resonances (peaks). This task can be tedious and time-consuming and often requires support by experienced users. Automated peak picking algorithms were introduced more than 25 years ago, but there are still major deficiencies/flaws that often prevent complete and error free peak picking of biological macromolecule spectra. The major challenges of automated peak picking algorithms is both the distinction of artifacts from real peaks particularly from those with irregular shapes and also picking peaks in spectral regions with overlapping resonances which are very hard to resolve by existing computer algorithms. In both of these cases a visual inspection approach could be more effective than a "blind" algorithm. We present a novel approach using computer vision (CV) methodology which could be better adapted to the problem of peak recognition. After suitable "training" we successfully applied the CV algorithm to spectra of medium sized soluble proteins up to molecular weights of 26 kDa and to a 130 kDa complex of a tetrameric membrane protein in detergent micelles. Our CV approach outperforms commonly used programs. With suitable training data sets the application of presented method can be extended to automated peak picking in multidimensional spectra of nucleic acids or carbohydrates and adapted to solid state NMR spectra. CV-Peak Picker is available upon request from the authors. © The Author (2015). Published by Oxford University Press. All rights reserved. For Permissions, please email:
    Bioinformatics 05/2015; 31(18). DOI:10.1093/bioinformatics/btv318 · 4.98 Impact Factor
  • Source
    • "Insomeapplicationssuchasobjecttracking,evenifthenumberofpossiblefeaturesisnotextensive,anexhaustive featureselectionisstillimpracticalduetocomputationalconstraints.In[93],theauthorsproposedagradientbased featureselectionschemeforonlineboostingwithprimaryapplicationsinpersondetectionandtracking.Theirwork iterativelyupdateseachfeatureusingagradientdescentalgorithm,byminimizingtheweightedleastsquareerror betweentheestimatedfeatureresponseandthetruelabel.Thisisparticularlyattractivefortrackingandupdating schemessuchas[82],whereatanytimeinstance,theobject'sappearanceisalreadyrepresentedbyaboostedclassifier learnedfrompreviousframes.Assumingthereisnodramaticchangeintheappearance,thegradientdescentbased algorithmcanrefinethefeaturesinaveryefficientmanner. Therehavealsobeenmanyfeaturesthatattemptedtomodeltheshapeoftheobjects.Forinstance,in[94]multi- pleboundaryfragmentstoweakclassifierswerecomposedandformedastrong"boundary-fragment-model"detector usingboosting.Theyensuredthefeasibilityofthefeatureselectionprocessbylimitingthenumberofboundary fragmentsto2-3foreachweakclassifier.In[95]theobjectdetectorswerelearnedwithaboostingalgorithmand thefeaturesetconsistedofarandomlychosendictionaryofcontourfragments.Averysimilaredgeletfeaturewas proposedin[96],andwasusedtolearnhumanbodypartdetectorsinordertohandlemultiple,partiallyoccludedhu- mans.In[97],shapeletfeaturesfocusingonlocalregionsoftheimagewerebuiltfromlow-levelgradientinformation usingAdaBoostforpedestriandetection.Aninterestingsidebenefitofhavingcontour/edgeletfeaturesisthatobject detectionandobjectsegmentationcanbeperformedjointly,suchastheworkin[98]and[99] "
    [Show abstract] [Hide abstract]
    ABSTRACT: Face detection is one of the most studied topics in computer vision literature, not only because of the challenging nature of face as an object, but also due to the countless applications that require the application of face detection as a first step. During the past 15 years, tremendous progress has been made due to the availability of data in unconstrained capture conditions (so-called ’in-the-wild’) through the Internet, the effort made by the community to develop publicly available benchmarks, as well as the progress in the development of robust computer vision algorithms. In this paper, we survey the recent advances in real-world face detection techniques, beginning with the seminal Viola-Jones face detector methodology. These techniques are roughly categorized into two general schemes: rigid templates, learned mainly via boosting based methods or by the application of deep neural networks, and deformable models that describe the face by its parts. Representative methods will be described in detail, along with a few additional successful methods that we briefly go through at the end. Finally, we survey the main databases used for the evaluation of face detection algorithms and recent benchmarking efforts, and discuss the future of face detection.
    Computer Vision and Image Understanding 04/2015; 138. DOI:10.1016/j.cviu.2015.03.015 · 1.54 Impact Factor
Show more


57 Reads
Available from