Scheduled System Maintenance:
On May 6th, single article purchases and IEEE account management will be unavailable from 8:00 AM - 12:00 PM ET (12:00 - 16:00 UTC). We apologize for the inconvenience.
By Topic

Computer and Robot Vision, 2009. CRV '09. Canadian Conference on

Date 25-27 May 2009

Filter Results

Displaying Results 1 - 25 of 53
  • [Front cover]

    Publication Year: 2009 , Page(s): C1
    Save to Project icon | Request Permissions | PDF file iconPDF (1021 KB)  
    Freely Available from IEEE
  • [Title page i]

    Publication Year: 2009 , Page(s): i
    Save to Project icon | Request Permissions | PDF file iconPDF (20 KB)  
    Freely Available from IEEE
  • [Title page iii]

    Publication Year: 2009 , Page(s): iii
    Save to Project icon | Request Permissions | PDF file iconPDF (66 KB)  
    Freely Available from IEEE
  • [Copyright notice]

    Publication Year: 2009 , Page(s): iv
    Save to Project icon | Request Permissions | PDF file iconPDF (104 KB)  
    Freely Available from IEEE
  • Table of contents

    Publication Year: 2009 , Page(s): v - viii
    Save to Project icon | Request Permissions | PDF file iconPDF (155 KB)  
    Freely Available from IEEE
  • Conference Information

    Publication Year: 2009 , Page(s): ix - xi
    Save to Project icon | Request Permissions | PDF file iconPDF (66 KB)  
    Freely Available from IEEE
  • Optical Flow from Motion Blurred Color Images

    Publication Year: 2009 , Page(s): 1 - 7
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (7065 KB) |  | HTML iconHTML  

    This paper presents an algorithm for the estimation of optical flow from a single, motion-blurred, color image. The proposed algorithm is based on earlier work that estimated the optical flow using the information from a single grey scale image. By treating the three color channels separately we improved the robustness of our approach. Since first introduced, different groups have used similar techniques of the original algorithm to estimate motion in a variety of applications. Experimental results from natural as well as artificially motion-blurred images are presented. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Multiple Hypothesis Tracking Method with Fragmentation Handling

    Publication Year: 2009 , Page(s): 8 - 15
    Cited by:  Papers (3)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (558 KB) |  | HTML iconHTML  

    In this paper, we present a new multiple hypotheses tracking (MHT) approach. Our tracking method is suitable for online applications, because it labels objects at every frame and estimates the best computed trajectories up to the current frame. In this work we address the problems of object merging and splitting (occlusions) and object fragmentations. Object fragmentation resulting from imperfect background subtraction can easily be confused with splitting objects in a scene, especially in close range surveillance applications. This subject is not addressed in most MHT methods. In this work, we propose a framework for MHT which distinguishes fragmentation and splitting using their spatial and temporal characteristics and by generating hypotheses only for splitting cases using observation in later frames. This approach results in a more accurate data association and a reduced size of the hypothesis graph. Our tracking method is evaluated with various indoor videos. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Efficient Target Recovery Using STAGE for Mean-shift Tracking

    Publication Year: 2009 , Page(s): 16 - 22
    Cited by:  Papers (3)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (622 KB) |  | HTML iconHTML  

    Robust visual tracking is a challenging problem, especially when a target undergoes complete occlusion or leaves and later re-enters the camera view. The mean-shift tracker is an efficient appearance-based tracking algorithm that has become very popular in recent years. Many researchers have developed extensions to the algorithm that improve the appearance model used in target localization. We approach the problem from a slightly different angle and seek to improve the robustness of the mean-shift tracker by integrating an efficient failure recovery mechanism. The proposed method uses a novel application of the STAGE algorithm to efficiently recover a target in the event of tracking failure. The STAGE algorithm boosts the performance of a local search algorithm by iteratively learning an evaluation function to predict good states for initiating searches. STAGE can be viewed as a random-restart algorithm that chooses promising restart states based on the shape of the state space, as estimated using the search trajectories from previous iterations. In the proposed method, an adapted version of STAGE is applied to the mean-shift target localization algorithm (Bhattacharyya coefficient maximization using the mean-shift procedure) to efficiently recover the lost target. Experiments indicate that the proposed method is viable as a technique for recovering from failure caused by complete occlusion or departure from the camera view. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Scalable Near-Optimal Recursive Structure from Motion

    Publication Year: 2009 , Page(s): 23 - 30
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (949 KB) |  | HTML iconHTML  

    Rao-BlackWellized particle filters have achieved a breakthrough in the scalability of filters used for Structure from Motion (SFM) and Simultaneous Localization And Mapping (SLAM). The new generations of these filters employ as proposal distribution the optimal i.e, the one taking into consideration not only the previous motion of the camera, but also the most recent measurement. However the way they sample from this importance function is not optimal since the locations of 3-d features are updated using a motion predicted only from the previous state. This results in a performance lower than the Extended Kalman Filters (EKF)s. We propose in this paper an approach that bears similarity with the Random Sample Consensus (RANSAC) paradigm and that enables us to sample more efficiently from the optimal importance function. It allows us to update the depth based on a motion updated using information from the most recent image and hence the updated samples would have a higher chance to be in regions corresponding to high posterior probability. This results in a performance equal to the performance of the EKF with much higher scalability. Also, our samples being generated and updated based on random sampling of the features, this provides an improved robustness to outliers. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Robot Control and Augmented Reality Interface for Multiple Robots

    Publication Year: 2009 , Page(s): 31 - 36
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (913 KB) |  | HTML iconHTML  

    Augmented reality technology is useful for interaction with robotic devices. A user can see the status of and direct orders to robots in the actual operating environment rather than on a remote computer console. A system is presented where fiducial markers are used for the dual purpose of reading the position of a multi-robot system and for user interaction using augmented reality. A user can observe, with the assistance of a hand-held device or wearable eyepiece display, the planned motion and tasks of the robots. An intuitive interface allows the robots to be retasked dynamically. A scalable system architecture and a prototype is presented. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Canine Pose Estimation: A Computing for Public Safety Solution

    Publication Year: 2009 , Page(s): 37 - 44
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (2709 KB) |  | HTML iconHTML  

    In this paper we discuss determining canine pose in the context of common poses observed in urban search and rescue dogs through the use a sensor network made up of accelerometers. We discuss the use of the canine pose estimation system in a disaster environment, and propose techniques for determining canine pose. In addition we discuss the challenges with this approach in such environments. This paper presents the experimental results obtained from the heavy urban search and rescue disaster simulation, where experiments were conducted using multiple canines, which show that angles can be derived from acceleration readings. Our experiments show that similar angles were measured for each of the poses, even when measured on multiple USAR canines of varying size. We also developed an algorithm to determine poses and display the current canine pose to the screen of a laptop. The algorithm was successful in determining some poses and had difficulty with others. These results are presented and discussed in this paper. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Generic Moment Invariants Based Supervised Learning Framework for Classification Using Partial Object Information

    Publication Year: 2009 , Page(s): 45 - 52
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1855 KB) |  | HTML iconHTML  

    We present a novel classification scheme which uses partial object information that is selected adaptively using modified distance transform and represented as moment invariants (Hu moments) to compensate for scale, translation and rotational transformation(s). The moment invariants of different parts of an object are learned using AdaBoost algorithm [1]. The classifier obtained using the proposed scheme is able to handle changes in illumination, pose, and varying inter-class and intra-class attributes. Partial information based classification shows robustness against object articulations, clutters, and occlusions. The first contribution of our proposed method is an adaptive selection of partial object information using modified distance transform that attempts to extract contours along with its neighborhood information in the form of blocks. Secondly, our proposed method is invariant to scaling, translation and rotation, and reliably classifies occluded objects using fractional information. Our proposed method achieved better detection and classification rate compared to other state-of-the-art schemes. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Probabilistic 3D Tracking: Rollator Users' Leg Pose from Coronal Images

    Publication Year: 2009 , Page(s): 53 - 60
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (5751 KB) |  | HTML iconHTML  

    Understanding the human gait is an important objective towards improving elderly mobility. In turn, gait analyses largely depend on kinematic and dynamic measurements. While the majority of current markerless vision systems focus on estimating 2D and 3D walking motion in the sagittal plane, we wish to estimate the 3D pose of rollator users' lower limbs from observing image sequences in the coronal (frontal) plane. Our apparatus poses a unique set of challenges: a single monocular view of only the lower limbs and a frontal perspective of the rollator user. Since motion in the coronal plane is relatively subtle, we explore multiple cues within a Bayesian probabilistic framework to formulate a posterior estimate for a given subject's leg limbs. This paper describes four cues based on three features to formulate a pose estimate: image gradients, colour and anthropometric symmetry. Our appearance model is applied within a non-parametric (particle) filtering system to track the lower limbs. Our tracking system does not rely on any detection for automatic initialization. Preliminary experiments are promising, showing that the algorithm may provide an indication of relative depth for each lower limb. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Bayesian Algorithm for Reading 1D Barcodes

    Publication Year: 2009 , Page(s): 61 - 67
    Cited by:  Papers (3)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (604 KB) |  | HTML iconHTML  

    The 1D barcode is a ubiquitous labeling technology, with symbologies such as UPC used to label approximately 99% of all packaged goods in the US. It would be very convenient for consumers to be able to read these barcodes using portable cameras (e.g. mobile phones), but the limited quality and resolution of images taken by these cameras often make it difficult to read the barcodes accurately. We propose a Bayesian framework for reading 1D barcodes that models the shape and appearance of barcodes, allowing for geometric distortions and image noise, and exploiting the redundant information contained in the parity digit. An important feature of our framework is that it doesn't require that every barcode edge be detected in the image. Experiments on a publicly available dataset of barcode images explore the range of images that are readable, and comparisons with two commercial readers demonstrate the superior performance of our algorithm. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Face Classification Using Gabor Wavelets and Random Forest

    Publication Year: 2009 , Page(s): 68 - 73
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (188 KB) |  | HTML iconHTML  

    This paper presents a new face classification technique based on Gabor wavelets and random forest. Random forest is a tree based classifier that consists of many decision trees. Each tree gives a classification and the output is the aggregate of these classifications. The proposed algorithm first extracts features from the face images using Gabor wavelet transform and then uses the random forest algorithm to classify the images based on the extracted features. But Gabor wavelet transform leads to high feature dimensions which increases the cost of computation. The proposed algorithm uses a random forest which selects a small set of most discriminant Gabor wavelet features. Only this small set of features is now used to classify the images resulting in a fast face recognition technique. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Video Pause Detection Using Wavelets

    Publication Year: 2009 , Page(s): 74 - 80
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1945 KB) |  | HTML iconHTML  

    As the volume of digital video captured and stored continues to increase, research efforts have focused on content management systems for video indexing and retrieval applications. A first step in generic video processing is shot boundary detection. This paper addresses a novel algorithm for abrupt shot (cut/pause) detection-especially on frames with similar statistics-based on the wavelet transform and content entropy. The algorithm has been successfully tested on some video categories including sport and live videos. Its quantitative performance has been compared to other known methods including pixel, histogram, frequency domain and statistics difference. In each test, the proposed wavelet method outperforms the others. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Near-Real-Time Image Matting with Known Background

    Publication Year: 2009 , Page(s): 81 - 87
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1734 KB) |  | HTML iconHTML  

    This paper presents a novel matting algorithm for handling cases where foreground objects are in front of complex (non-smooth) but known background. The algorithm is based on a Poisson equation that takes the gradient of the known background into consideration. The quantitative evaluation based on ground truth shows that the matting results obtained using the proposed algorithm is more accurate than the ones generated by the global Poisson matting. The proposed algorithm is also optimized for parallel processing and runs in near-real-time on programmable graphics hardware. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Motion Histogram Analysis Based Key Frame Extraction for Human Action/Activity Representation

    Publication Year: 2009 , Page(s): 88 - 92
    Cited by:  Papers (3)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (3970 KB) |  | HTML iconHTML  

    Key frame extraction is an important technique in video summarization, browsing, searching, and understanding. In this paper, a novel algorithm for key frame extraction based on intra-frame and inter-frame motion histogram analysis is proposed. The extracted key frames contain complex motion and are salient in respect to their neighboring frames, and can be used to represent actions and activities in video. The key frames are first initialized by finding peaks in the curve of entropy calculated on motion histograms in each video frame. The peaked entropies are then weighted by inter-frame saliency which we use histogram intersection to output final key frames. The effectiveness of the proposed method is validated by a large variety of real-life videos. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Towards Navigation Summaries: Automated Production of a Synopsis of a Robot Trajectories

    Publication Year: 2009 , Page(s): 93 - 100
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (5934 KB) |  | HTML iconHTML  

    In this paper we describe an approach to computing a navigation summary:a visual synopsis of the notable images that characterize a trajectory. We use a combination of PCA and supplementary features to ensure both converge of the trajectory and appearance spaces. The results obtained from a series of experiments are promising and provide us with a method to index and classify video footage from our robot. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • JEDI: Adaptive Stochastic Estimation for Joint Enhancement and Despeckling of Images for SAR

    Publication Year: 2009 , Page(s): 101 - 107
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (971 KB) |  | HTML iconHTML  

    Synthetic aperture radar (SAR) images are degraded by a form of multiplicative noise known as speckle. Current methods for despeckling are limited in that they either do not perform enough noise attenuation, or do not adequately preserve or enhance image detail. We propose a novel adaptive stochastic method for joint enhancement and despecking of images (JEDI) for SAR. The proposed method utilizes an adaptive importance sampling scheme based on local statistics to generate random samples while reducing estimation variance. A Monte Carlo estimate is computed based on the generated samples, wherein the samples are aggregated to form a despeckled and detail-enhanced result. The advantage of JEDI is the ability to efficiently take advantage of information redundancy in speckled images to reduce the effects of speckle while simultaneously enhancing detail visualization. Testing with both simulated and real speckled images shows that JEDI typically outperforms popular despeckling algorithms such as Frost filtering, anisotropic diffusion, median filtering, Gamma-MAP and GenLik in terms of quantitative and qualitative visual quality. On average, JEDI provides a 2-15% improvement in PSNR and a 5-14% improvement in image quality index measures over the tested methods. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive Monte Carlo Retinex Method for Illumination and Reflectance Separation and Color Image Enhancement

    Publication Year: 2009 , Page(s): 108 - 115
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (751 KB) |  | HTML iconHTML  

    A novel stochastic Retinex method based on adaptive Monte Carlo estimation is presented for the purpose of illumination and reflectance separation and color image enhancement. A spatially-adaptive sampling scheme is employed to generate a set of random samples from the image field. A Monte Carlo estimate of the illumination is computed based on the Pearson Type VII error statistics of the drawn samples. The proposed method takes advantage of both local and global contrast information to provide better separation of reflectance and illumination by reducing the effects of strong shadows and other sharp illumination changes on the estimation process, improving the preservation of the original photographic tone, and avoiding the amplification of noise in dark regions. Experimental results using monochromatic face images under different illumination conditions and low-contrast chromatic images show the effectiveness of the proposed method for illumination and reflectance separation and color image enhancement when compared to existing Retinex and color enhancement techniques. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Non-Accidental Features for Gesture Spotting

    Publication Year: 2009 , Page(s): 116 - 123
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (798 KB) |  | HTML iconHTML  

    In this paper we argue that gestures based on non-accidental motion features can be reliably detected amongst unconstrained background motion. Specifically, we demonstrate that humans can perform non-accidental motions with high accuracy, and that these trajectories can be extracted from video with sufficient accuracy to reliably distinguish them from the background motion. We demonstrate this by learning Gaussian mixture models of the features associated with gesture. Non-accidental features result in compact, heavily-weighted, mixture component distributions. We demonstrate reliable detection by using the mixture models to discriminate non-accidental features from the background. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • An Efficient and Fast Active Contour Model for Salient Object Detection

    Publication Year: 2009 , Page(s): 124 - 131
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (6091 KB) |  | HTML iconHTML  

    In this paper, we investigate the polarity information to improve the active contour model proposed by Chunming et al. Unlike the traditional level set formulations, the variational level set formulation proposed by forces the level set function to be close to a signed distance function,and therefore completely eliminates the need of the reinitialization procedure and speeds up the curve evolution.However, like the majority of classical active contour models,the model proposed by relies on a gradient based stopping function, depending on the image gradient, to stop the curve evolution. Consequently, using gradient information for noisy and textured images, the evolving curve may pass through or stop far from the salient object boundaries.Moreover, in this case, the isotropic smoothing Gaussian has to be strong, which will smooth the edges too. For these reasons, we propose the use of a polarity based stopping function. In fact, comparatively to the gradient information,the polarity information accurately distinguishes the boundaries or edges of the salient objects. Hence, combining the polarity information with the active contour model of we obtain a fast and efficient active contour model for salient object detection. Experiments are performed on several images to show the advantage of the polarity based active contour. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Vision Based Metal Spectral Analysis Using Multi-label Classification

    Publication Year: 2009 , Page(s): 132 - 139
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (481 KB) |  | HTML iconHTML  

    Industrial equipments that employ element identification tend to be expensive as they utilize built-in spectroscopes and computers for post processing. In this paper we present an in situ fully automatic method for detecting constituent elements in a sample specimen using computer vision and machine learning techniques on Laser Induced Breakdown Spectroscopy (LIBS) spectra. This enables the development of a compact and portable spectrometer on a high resolution video camera. In the traditional classification problem, classes are mutually exclusive by definition. However, in spectral analysis a spectrum could contain emissions from multiple elements such that the disjointness of the labels is no longer valid. We cast the metal detection problem as a multi-label classification and enable detection of elemental composition of the specimen. Here, we apply both Support Vector Machine (SVM) and Artificial Neural Networks (ANN) to multiple metal classification and compare the performance with a simple template matching technique. Both machine learning approaches yield correct identification of metals to an accuracy of 99%. Our method is useful in instances where accurate elemental analysis is not required but rather a qualitative analysis. Experiments on the simulation data show that our method is suitable for LIBS metal detection. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.