Raluca Brehar

The influence of hubness on nearest-neighbor methods in object recognition

Object recognition from images is one of the essential problems in automatic image processing. In this paper we focus specifically on nearest neighbor methods, which are widely used in many practical applications, not necessarily related to image data. It has recently come to attention that high dimensional data also exhibit high hubness, which essentially means that some very influential data points appear and these points are referred to as hubs. Unsurprisingly, hubs play a very important role in the nearest neighbor classification. We examine the hubness of various image data sets, under several different feature representations. We also show that it is possible to exploit the observed hubness and improve the recognition accuracy.

Pedestrians detection using a cascade of LBP and HOG classifiers

Accurate pedestrian detection in urban environment is a highly explored research field. We propose a new approach in pedestrian detection that combines the popular Local Binary Patterns and Histogram of Oriented Gradient features. The novelty of our work resides in the combination of a reduced HOG feature vector with uniform LBP patterns for the pedestrian data representation. Another contribution resides in the design and implementation of a two-stage cascade classifier of Support Vector Machine. Our method has been trained and tested on reference benchmark datasets and it proved to have good results.

Pedestrian detection in infrared images using Aggregated Channel Features

We propose a method for detecting pedestrians in infrared images. The method combines a fast region of interest generator with fast feature pyramid object detection. Knowing the appearance model of pedestrians in infrared images we infer some edge and intensity based filters that generate the regions in which pedestrian hypotheses may appear. On those regions we apply the Aggregated Channel Features introduced by [1]. We train and test the proposed solution on an infrared pedestrian data set and the results show a good detection accuracy and small execution time of about 30fps.

Scan window based pedestrian recognition methods improvement by search space and scale reduction

Most of computation time when dealing with a pedestrian detector is spent in the feature computation and then in the multi-scale classification. This second step consists of applying scanning windows at multiple scales. Depending on the number of scales and on the image dimension, this step is slow because a large number of windows is generated. An efficient pruning algorithm able to remove most of the scan windows brings a significant contribution on the overall execution time. We propose a scan window pruning algorithm based on a combination of several filters: (1) remove windows based on the relation between the dimension and the position in the scene; (2) remove uniform regions from the image such as the sky or the road; (3) remove regions with high density of horizontal edges such as vegetation parts; (4) keep local maxima windows having a high density of connected vertical edges. The combination of these four filters eliminates more than 90% of the scanning windows in a given image and maintains the windows that are overlapped on regions with a high probability of representing a pedestrian. We have integrated our method in a generic framework for pedestrian detection [1] and we have studied two aspects: (1)how the performance of the algorithms varies with respect to the filters proposed by our pruning strategy and (2) what is the speed gain when the quality loss is negligible. The proposed filters have a negligible loss in performance and even improve it in some cases while the execution time is improved.

Part-based pedestrian detection using HoG features and vertical symmetry

This paper describes a new approach for pedestrian detection in traffic scenes. The originality of the method resides in the combination of the benefits of the symmetry characteristic for pedestrians in intensity images and the benefits of deformable part-based models for recognizing pedestrians in multiple object hypotheses generated by a stereo vision system. A mixture model based on several pedestrian attitudes is used for addressing the large intraclass variability that pedestrians may have (they may have different poses and attitudes like: standing, walking, running etc). We have used a probabilistic approach based on support vector machine (SVM) and histograms of gradient orientations (HoG) features for pedestrian classification.

Pedestrian Detection in Infrared Images Using HOG, LBP, Gradient Magnitude and Intensity Feature Channels

Recent work in monocular pedestrian detection is trying to improve the execution time while keeping the accuracy as high as possible. A popular and successful approach for monocular intensity pedestrian detection is based on the approximation (instead of computation) of image features for multiple scales based on the features computed on set of predefined scales. We port this idea to the infrared domain. Our contributions reside in the combination of four channel features, namely infrared, histogram of gradient orientations, normalized gradient magnitude and local binary patterns with the objective of detecting pedestrians for night vision applications dealing with far infrared sensors. Multiple scale feature computation is done by feature approximation. Another contribution is the study of different formulations for Local Binary Patterns like uniform patterns and rotation invariant patterns and their effect on detection performance. The detection speed is also boosted by the aid of a fast morphological based region of interest generator. We vary the number of approximated scales per octave and study the impact on execution time and accuracy. A reasonable result hits a speed of 18fps with a log average miss rate of 39%.

A comparative study of pedestrian detection methods using classical Haar and HoG features versus bag of words model computed from Haar and HoG features

The bag of words model has been actively adopted by content based image retrieval and image annotation techniques. We employ this model for the particular task of pedestrian detection in two dimensional images, producing this way a novel approach to pedestrian detection. The experiments we have done in this paper compare the behavior of discriminative recognition approaches that use AdaBoost on codebook features versus Adaboost trained on primitive features that may be extracted from a two dimensional image. By primitive features we refer in this paper to Haar features and Histogram of Oriented Gradients both being extremely used in object recognition in general and in pedestrian detection in particular. The conclusion of our experiments is that the codebook representation performs better than the primitive feature representation.

Spatio-temporal reasoning for traffic scene understanding

In this paper we introduce a system for semantic understanding of traffic scenes. The system detects objects in video images captured in real vehicular traffic situations, classifies them, maps them to the OpenCyc1 ontology and finally generates descriptions of the traffic scene in CycL or cvasi-natural language. We employ meta-classification methods based on AdaBoost and Random forest algorithms for identifying interest objects like: cars, pedestrians, poles in traffic and we derive a set of annotations for each traffic scene. These annotations are mapped to OpenCyc concepts and predicates, spatiotemporal rules for object classification and scene understanding are then asserted in the knowledge base. Finally, we show that the system performs well in understanding traffic scene situations and summarizing them. The novelty of the approach resides in the combination of stereo-based object detection and recognition methods with logic based spatio-temporal reasoning.

Local information statistics of LBP and HOG for pedestrian detection

We present several methods of pedestrian detection in intensity images using different local statistical measures applied to two classes of features extensively used in pedestrian detection: uniform local binary patterns - LBP and a modified version of histogram of oriented gradients - HOG. Our work extracts local binary patterns and magnitude and orientation of the gradient image. Then we divide the image into blocks. Within each block we extract different statistics like: histogram (weighted by the gradient magnitude in the case of HOG), information, entropy and energy of the local binary code. We use Adaboost for training four classifiers and we analyze the classification error of each method on the Daimler benchmark pedestrian dataset.

Object recognition in wikimage data based on local invariant image features

Object recognition is an essential task in content-based image retrieval and classification. This paper deals with object recognition in WIKImage data, a collection of publicly available annotated Wikipedia images. WIKImage comprises a set of 14 binary classification problems with significant class imbalance. Our approach is based on using the local invariant image features and we have compared 3 standard and widely used feature types: SIFT, SURF and ORB. We have examined how the choice of representation affects the k-nearest neighbor data topology and have shown that some feature types might be more appropriate than others for this particular problem. In order to assess the difficulty of the data, we have evaluated 7 different k-nearest neighbor classification methods and shown that the recently proposed hubness-aware classifiers might be used to either increase the accuracy of prediction, or the macro-averaged F-score. However, our results indicate that further improvements are possible and that including the textual feature information might prove beneficial for system performance.