|
|
Point and Line Parameterizations Using Parallel Coordinates for Hough Transform
Juránková, Markéta ; Kälviäinen, Heikki (oponent) ; Kittler, Josef (oponent) ; Herout, Adam (vedoucí práce)
This thesis focuses on usage of parallel coordinates for line and point parameterizations. The parallel coordinate system represents the space with axes which are mutually parallel. A point from two-dimensional Euclidean space is in parallel coordinates represented by a line and a line is represented by a point. This property can by used for the Hough transform - a method, where the points of interest vote in parameter space for possible hypotheses. Parameterizations by the parallel coordinates require only rasterization of lines, therefore it is very fast and accurate. In this thesis, the parameterizations are used for matrix code and vanishing points detection.
|
|
|
Acceleration of Object Detection Using Classifiers
Juránek, Roman ; Kälviäinen, Heikki (oponent) ; Sojka, Eduard (oponent) ; Zemčík, Pavel (vedoucí práce)
Detection of objects in computer vision is a complex task. One of most popular and well explored approaches is the use of statistical classifiers and scanning windows. In this approach, classifiers learned by AdaBoost algorithm (or some modification) are often used as they achieve low error rates, high detection rates and they are suitable for detection in real-time applications. Object detection run-time which uses such classifiers can be implemented by various methods and properties of underlying architecture can be used for speed-up of the detection. For the purpose of acceleration, graphics hardware, multi-core architectures, SIMD or other means can be used. The detection is often implemented on programmable hardware. The contribution of this thesis is to introduce an optimization technique which enhances object detection performance with respect to an user defined cost function. The optimization balances computations of previously learned classifiers between two or more run-time implementations in order to minimize the cost function. The optimization method is verified on a basic example -- division of a classifier to a pre-processing unit implemented in FPGA, and a post-processing unit in standard PC.
|
|
|
Point to Line Mappings and Other Line Parameterizations not only for Hough Transform
Havel, Jiří ; Kälviäinen, Heikki (oponent) ; Lefevre, Sebastien (oponent) ; Herout, Adam (vedoucí práce)
This works focuses on the Hough transform (HT). The HT is mostly used for the detection of lines or curves, but was also generalized for detection of arbitrary shapes. The main theme of this work are line parameterizations, especially the Point-to-Line mappings. These parameterizations share the property, that a point in the image maps onto a line in the parameter space. This work presents proofs of some properties of PTLMs, notably the existence of a practical pair of PTLMs for line detection and the effect of a convolution in the image space on the contents of the parameter space. Two realtime implementations of HT are presented in this work. Both accelerate HT using graphical hardware. One uses GPGPU API CUDA and the other the rendering API OpenGL. As an application of the line detection, this work describes part of the detection of checkerboard marker usable for the augmented reality.
|
|
|
Sharing Local Information for Faster Scanning-Window Object Detection
Hradiš, Michal ; Kälviäinen, Heikki (oponent) ; Matas, Jiří (oponent) ; Zemčík, Pavel (vedoucí práce)
This thesis aims to improve existing scanning-window object detectors by exploiting information shared among neighboring image windows. This goal is realized by two novel methods which are build on the ideas of Wald's Sequential Probability Ratio Test and WaldBoost. Early non-Maxima Suppression moves non-maxima suppression decisions from a post-processing step to an early classification phase in order to make the decisions as soon as possible and thus avoid normally wasted computations. Neighborhood suppression enhances existing detectors with an ability to suppress evaluation at overlapping positions. The proposed methods are applicable to a wide range of detectors. Experiments show that both methods provide significantly better speed-precision trade-off compared to state-of-the-art WaldBoost detectors which process image windows independently. Additionally, the thesis presents results of extensive experiments which evaluate commonly used image features in several detection tasks and scenarios.
|
|
|
Lifting Scheme Cores for Wavelet Transform
Bařina, David ; Kälviäinen, Heikki (oponent) ; Sojka, Eduard (oponent) ; Zemčík, Pavel (vedoucí práce)
The thesis focuses on efficient computation of the two-dimensional discrete wavelet transform. The state-of-the-art methods are extended in several ways to perform the transform in a single loop, possibly in multi-scale fashion, using a compact streaming core. This core can further be appropriately reorganized to target the minimization of certain platform resources. The approach presented here nicely fits into common SIMD extensions, exploits the cache hierarchy of modern general-purpose processors, and is suitable for parallel evaluation. Finally, the approach presented is incorporated into the JPEG 2000 compression chain, in which it has proved to be fundamentally faster than widely used implementations.
|
|
|
Acceleration of Object Detection Using Classifiers
Juránek, Roman ; Kälviäinen, Heikki (oponent) ; Sojka, Eduard (oponent) ; Zemčík, Pavel (vedoucí práce)
Detection of objects in computer vision is a complex task. One of most popular and well explored approaches is the use of statistical classifiers and scanning windows. In this approach, classifiers learned by AdaBoost algorithm (or some modification) are often used as they achieve low error rates, high detection rates and they are suitable for detection in real-time applications. Object detection run-time which uses such classifiers can be implemented by various methods and properties of underlying architecture can be used for speed-up of the detection. For the purpose of acceleration, graphics hardware, multi-core architectures, SIMD or other means can be used. The detection is often implemented on programmable hardware. The contribution of this thesis is to introduce an optimization technique which enhances object detection performance with respect to an user defined cost function. The optimization balances computations of previously learned classifiers between two or more run-time implementations in order to minimize the cost function. The optimization method is verified on a basic example -- division of a classifier to a pre-processing unit implemented in FPGA, and a post-processing unit in standard PC.
|
|
|
Point to Line Mappings and Other Line Parameterizations not only for Hough Transform
Havel, Jiří ; Kälviäinen, Heikki (oponent) ; Lefevre, Sebastien (oponent) ; Herout, Adam (vedoucí práce)
This works focuses on the Hough transform (HT). The HT is mostly used for the detection of lines or curves, but was also generalized for detection of arbitrary shapes. The main theme of this work are line parameterizations, especially the Point-to-Line mappings. These parameterizations share the property, that a point in the image maps onto a line in the parameter space. This work presents proofs of some properties of PTLMs, notably the existence of a practical pair of PTLMs for line detection and the effect of a convolution in the image space on the contents of the parameter space. Two realtime implementations of HT are presented in this work. Both accelerate HT using graphical hardware. One uses GPGPU API CUDA and the other the rendering API OpenGL. As an application of the line detection, this work describes part of the detection of checkerboard marker usable for the augmented reality.
|
|
|
Sharing Local Information for Faster Scanning-Window Object Detection
Hradiš, Michal ; Kälviäinen, Heikki (oponent) ; Matas, Jiří (oponent) ; Zemčík, Pavel (vedoucí práce)
This thesis aims to improve existing scanning-window object detectors by exploiting information shared among neighboring image windows. This goal is realized by two novel methods which are build on the ideas of Wald's Sequential Probability Ratio Test and WaldBoost. Early non-Maxima Suppression moves non-maxima suppression decisions from a post-processing step to an early classification phase in order to make the decisions as soon as possible and thus avoid normally wasted computations. Neighborhood suppression enhances existing detectors with an ability to suppress evaluation at overlapping positions. The proposed methods are applicable to a wide range of detectors. Experiments show that both methods provide significantly better speed-precision trade-off compared to state-of-the-art WaldBoost detectors which process image windows independently. Additionally, the thesis presents results of extensive experiments which evaluate commonly used image features in several detection tasks and scenarios.
|
|
|
Point and Line Parameterizations Using Parallel Coordinates for Hough Transform
Juránková, Markéta ; Kälviäinen, Heikki (oponent) ; Kittler, Josef (oponent) ; Herout, Adam (vedoucí práce)
This thesis focuses on usage of parallel coordinates for line and point parameterizations. The parallel coordinate system represents the space with axes which are mutually parallel. A point from two-dimensional Euclidean space is in parallel coordinates represented by a line and a line is represented by a point. This property can by used for the Hough transform - a method, where the points of interest vote in parameter space for possible hypotheses. Parameterizations by the parallel coordinates require only rasterization of lines, therefore it is very fast and accurate. In this thesis, the parameterizations are used for matrix code and vanishing points detection.
|
|
|
Lifting Scheme Cores for Wavelet Transform
Bařina, David ; Kälviäinen, Heikki (oponent) ; Sojka, Eduard (oponent) ; Zemčík, Pavel (vedoucí práce)
The thesis focuses on efficient computation of the two-dimensional discrete wavelet transform. The state-of-the-art methods are extended in several ways to perform the transform in a single loop, possibly in multi-scale fashion, using a compact streaming core. This core can further be appropriately reorganized to target the minimization of certain platform resources. The approach presented here nicely fits into common SIMD extensions, exploits the cache hierarchy of modern general-purpose processors, and is suitable for parallel evaluation. Finally, the approach presented is incorporated into the JPEG 2000 compression chain, in which it has proved to be fundamentally faster than widely used implementations.
|
host ::
RSS.