TESES E DISSERTAÇÕES
2016 |
Gabriel Resende Goncalves License Plate Recognition based on Temporal Redundancy Masters Thesis Federal University of Minas Gerais, 2016. Resumo | Links | BibTeX | Tags: Automatic License Plate Recognition, DeepEyes, GigaFrames @mastersthesis{Goncalves:2016:MSc, title = {License Plate Recognition based on Temporal Redundancy}, author = {Gabriel Resende Goncalves}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/03/dissertation_2016_Gabriel.pdf}, year = {2016}, date = {2016-08-26}, school = {Federal University of Minas Gerais}, abstract = {Recognition of vehicle license plates is an important task applied to a myriad of real scenarios. Most approaches in the literature first detect an on-track vehicle, locate the license plate, perform a segmentation of its characters and then recognize the characters using an Optical Character Recognition (OCR) approach. However, these approaches focus on performing these tasks using only a single frame of each vehicle in the video. Therefore, such techniques might have their recognition rates reduced due to noise present in that particular frame. On the other hand, in this work we propose an approach to automatically detect the vehicle on the road and identify (locate/recognize) its license plate based on temporal redundant information instead of selecting a single frame to perform the recognition. We also propose two post-processing steps that can be employed to improve the accuracy of the system by querying a license plate database (e.g., the Department of Motor Vehicles database containing a list of all issued license plates and car models). Experimental results demonstrate that it is possible to improve the vehicle recognition rate in 15.5 percentage points (p.p.) (an increase of 23.38%) of the baseline results, using our proposal temporal redundancy approach. Furthermore, additional 7.8 p.p. are achieved using the two post-processing approaches, leading to a final recognition rate of 89.6% on a dataset with 5,200 frame images of $300$ vehicles recorded at Federal University of Minas Gerais (UFMG). In addition, this work also proposes a novel benchmark, designed specifically to evaluate character segmentation techniques, composed of a dataset of 2,000 Brazilian license plates (resulting in 14,000 alphanumeric symbols) and an evaluation protocol considering a novel evaluation measure, the Jaccard-Centroid coefficient.}, keywords = {Automatic License Plate Recognition, DeepEyes, GigaFrames}, pubstate = {published}, tppubtype = {mastersthesis} } Recognition of vehicle license plates is an important task applied to a myriad of real scenarios. Most approaches in the literature first detect an on-track vehicle, locate the license plate, perform a segmentation of its characters and then recognize the characters using an Optical Character Recognition (OCR) approach. However, these approaches focus on performing these tasks using only a single frame of each vehicle in the video. Therefore, such techniques might have their recognition rates reduced due to noise present in that particular frame. On the other hand, in this work we propose an approach to automatically detect the vehicle on the road and identify (locate/recognize) its license plate based on temporal redundant information instead of selecting a single frame to perform the recognition. We also propose two post-processing steps that can be employed to improve the accuracy of the system by querying a license plate database (e.g., the Department of Motor Vehicles database containing a list of all issued license plates and car models). Experimental results demonstrate that it is possible to improve the vehicle recognition rate in 15.5 percentage points (p.p.) (an increase of 23.38%) of the baseline results, using our proposal temporal redundancy approach. Furthermore, additional 7.8 p.p. are achieved using the two post-processing approaches, leading to a final recognition rate of 89.6% on a dataset with 5,200 frame images of $300$ vehicles recorded at Federal University of Minas Gerais (UFMG). In addition, this work also proposes a novel benchmark, designed specifically to evaluate character segmentation techniques, composed of a dataset of 2,000 Brazilian license plates (resulting in 14,000 alphanumeric symbols) and an evaluation protocol considering a novel evaluation measure, the Jaccard-Centroid coefficient. |
Artur Jordao The Good, the Fast and the Better Pedestrian Detector Masters Thesis Federal University of Minas Gerais, 2016. Resumo | Links | BibTeX | Tags: DeepEyes, DET, GigaFrames, Pedestrian Detection, VER+ @mastersthesis{Jordao:2016:MSc, title = {The Good, the Fast and the Better Pedestrian Detector}, author = {Artur Jordao}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/03/dissertation_2016_ArturJordao.pdf}, year = {2016}, date = {2016-06-24}, school = {Federal University of Minas Gerais}, abstract = {Pedestrian detection is a well-known problem in Computer Vision, mostly because of its direct applications in surveillance, transit safety and robotics. In the past decade, several efforts have been performed to improve the detection in terms of accuracy, speed and feature enhancement. In this work, we propose and analyze techniques focusing on these points. First, we develop an accurate oblique random forest (oRF) associated with Partial Least Squares (PLS). The method utilizes the PLS to find a decision surface, at each node of a decision tree, that better splits the samples presented to it, based on some purity criterion. To measure the advantages provided by PLS on the oRF, we compare the proposed method with the oRF based on SVM. Second, we evaluate and compare filtering approaches to reduce the search space and keep only potential regions of interest to be presented to detectors, speeding up the detection process. Experimental results demonstrate that the evaluated filters are able to discard a large number of detection windows without compromising the accuracy. Finally, we propose a novel approach to extract powerful features regarding the scene. The method combines results of distinct pedestrian detectors by reinforcing the human hypothesis, whereas suppressing a significant number of false positives due to the lack of spatial consensus when multiple detectors are considered. Our proposed approach, referred to as Spatial Consensus (SC), outperforms all previously published state-of-the-art pedestrian detection methods.}, keywords = {DeepEyes, DET, GigaFrames, Pedestrian Detection, VER+}, pubstate = {published}, tppubtype = {mastersthesis} } Pedestrian detection is a well-known problem in Computer Vision, mostly because of its direct applications in surveillance, transit safety and robotics. In the past decade, several efforts have been performed to improve the detection in terms of accuracy, speed and feature enhancement. In this work, we propose and analyze techniques focusing on these points. First, we develop an accurate oblique random forest (oRF) associated with Partial Least Squares (PLS). The method utilizes the PLS to find a decision surface, at each node of a decision tree, that better splits the samples presented to it, based on some purity criterion. To measure the advantages provided by PLS on the oRF, we compare the proposed method with the oRF based on SVM. Second, we evaluate and compare filtering approaches to reduce the search space and keep only potential regions of interest to be presented to detectors, speeding up the detection process. Experimental results demonstrate that the evaluated filters are able to discard a large number of detection windows without compromising the accuracy. Finally, we propose a novel approach to extract powerful features regarding the scene. The method combines results of distinct pedestrian detectors by reinforcing the human hypothesis, whereas suppressing a significant number of false positives due to the lack of spatial consensus when multiple detectors are considered. Our proposed approach, referred to as Spatial Consensus (SC), outperforms all previously published state-of-the-art pedestrian detection methods. |
Cristianne Rodrigues Santos Dutra Técnicas Otimizadas para Reidentificaçâo de Pessoas Masters Thesis Federal University of Minas Gerais, 2016. Links | BibTeX | Tags: DeepEyes, GigaFrames, Person Re-Identification, VER+ @mastersthesis{Dutra:2016:MSc, title = {Técnicas Otimizadas para Reidentificaçâo de Pessoas}, author = {Cristianne Rodrigues Santos Dutra}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/02/thesis_2016_Cristianne.pdf}, year = {2016}, date = {2016-01-01}, school = {Federal University of Minas Gerais}, keywords = {DeepEyes, GigaFrames, Person Re-Identification, VER+}, pubstate = {published}, tppubtype = {mastersthesis} } |
2015 |
Cassio Elias Santos dos Junior Partial Least Squares for Face Hashing Masters Thesis Federal University of Minas Gerais, 2015. Resumo | Links | BibTeX | Tags: DeepEyes, Face Identification, Face Recognition, GigaFrames, Indexing Structure, Local Sensitive Hashing, Partial Least Squares, VER+ @mastersthesis{Santos:2015:MSc, title = {Partial Least Squares for Face Hashing}, author = {Cassio Elias Santos dos Junior}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/02/dissertation_2015_Cassio.pdf}, year = {2015}, date = {2015-08-24}, school = {Federal University of Minas Gerais}, abstract = {Face identification is an important research topic due to areas such as its application to surveillance, forensics and human-computer interaction. In the past few years, a myriad of methods for face identification has been proposed in the literature, with just a few among them focusing on scalability. In this work, we propose a simple but efficient approach for scalable face identification based on partial least squares (PLS) and random independent hash functions inspired by locality-sensitive hashing (LSH), resulting in the PLS for hashing (PLSH) approach. The original PLSH approach is further extended using feature selection to reduce the computational cost to evaluate the PLS-based hash functions, resulting in the state-of-the-art extended PLSH approach (ePLSH). The proposed approach is evaluated in the dataset FERET and in the dataset FRGCv1. The results show significant reduction in the number of subjects evaluated in the face identification (reduced to 0.3% of the gallery), providing averaged speedups up to 233 times compared to evaluating all subjects in the face gallery and 58 times compared to previous works in the literature.}, keywords = {DeepEyes, Face Identification, Face Recognition, GigaFrames, Indexing Structure, Local Sensitive Hashing, Partial Least Squares, VER+}, pubstate = {published}, tppubtype = {mastersthesis} } Face identification is an important research topic due to areas such as its application to surveillance, forensics and human-computer interaction. In the past few years, a myriad of methods for face identification has been proposed in the literature, with just a few among them focusing on scalability. In this work, we propose a simple but efficient approach for scalable face identification based on partial least squares (PLS) and random independent hash functions inspired by locality-sensitive hashing (LSH), resulting in the PLS for hashing (PLSH) approach. The original PLSH approach is further extended using feature selection to reduce the computational cost to evaluate the PLS-based hash functions, resulting in the state-of-the-art extended PLSH approach (ePLSH). The proposed approach is evaluated in the dataset FERET and in the dataset FRGCv1. The results show significant reduction in the number of subjects evaluated in the face identification (reduced to 0.3% of the gallery), providing averaged speedups up to 233 times compared to evaluating all subjects in the face gallery and 58 times compared to previous works in the literature. |
2016 |
Gabriel Resende Goncalves License Plate Recognition based on Temporal Redundancy Masters Thesis Federal University of Minas Gerais, 2016. Resumo | Links | BibTeX | Tags: Automatic License Plate Recognition, DeepEyes, GigaFrames @mastersthesis{Goncalves:2016:MSc, title = {License Plate Recognition based on Temporal Redundancy}, author = {Gabriel Resende Goncalves}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/03/dissertation_2016_Gabriel.pdf}, year = {2016}, date = {2016-08-26}, school = {Federal University of Minas Gerais}, abstract = {Recognition of vehicle license plates is an important task applied to a myriad of real scenarios. Most approaches in the literature first detect an on-track vehicle, locate the license plate, perform a segmentation of its characters and then recognize the characters using an Optical Character Recognition (OCR) approach. However, these approaches focus on performing these tasks using only a single frame of each vehicle in the video. Therefore, such techniques might have their recognition rates reduced due to noise present in that particular frame. On the other hand, in this work we propose an approach to automatically detect the vehicle on the road and identify (locate/recognize) its license plate based on temporal redundant information instead of selecting a single frame to perform the recognition. We also propose two post-processing steps that can be employed to improve the accuracy of the system by querying a license plate database (e.g., the Department of Motor Vehicles database containing a list of all issued license plates and car models). Experimental results demonstrate that it is possible to improve the vehicle recognition rate in 15.5 percentage points (p.p.) (an increase of 23.38%) of the baseline results, using our proposal temporal redundancy approach. Furthermore, additional 7.8 p.p. are achieved using the two post-processing approaches, leading to a final recognition rate of 89.6% on a dataset with 5,200 frame images of $300$ vehicles recorded at Federal University of Minas Gerais (UFMG). In addition, this work also proposes a novel benchmark, designed specifically to evaluate character segmentation techniques, composed of a dataset of 2,000 Brazilian license plates (resulting in 14,000 alphanumeric symbols) and an evaluation protocol considering a novel evaluation measure, the Jaccard-Centroid coefficient.}, keywords = {Automatic License Plate Recognition, DeepEyes, GigaFrames}, pubstate = {published}, tppubtype = {mastersthesis} } Recognition of vehicle license plates is an important task applied to a myriad of real scenarios. Most approaches in the literature first detect an on-track vehicle, locate the license plate, perform a segmentation of its characters and then recognize the characters using an Optical Character Recognition (OCR) approach. However, these approaches focus on performing these tasks using only a single frame of each vehicle in the video. Therefore, such techniques might have their recognition rates reduced due to noise present in that particular frame. On the other hand, in this work we propose an approach to automatically detect the vehicle on the road and identify (locate/recognize) its license plate based on temporal redundant information instead of selecting a single frame to perform the recognition. We also propose two post-processing steps that can be employed to improve the accuracy of the system by querying a license plate database (e.g., the Department of Motor Vehicles database containing a list of all issued license plates and car models). Experimental results demonstrate that it is possible to improve the vehicle recognition rate in 15.5 percentage points (p.p.) (an increase of 23.38%) of the baseline results, using our proposal temporal redundancy approach. Furthermore, additional 7.8 p.p. are achieved using the two post-processing approaches, leading to a final recognition rate of 89.6% on a dataset with 5,200 frame images of $300$ vehicles recorded at Federal University of Minas Gerais (UFMG). In addition, this work also proposes a novel benchmark, designed specifically to evaluate character segmentation techniques, composed of a dataset of 2,000 Brazilian license plates (resulting in 14,000 alphanumeric symbols) and an evaluation protocol considering a novel evaluation measure, the Jaccard-Centroid coefficient. |
Artur Jordao The Good, the Fast and the Better Pedestrian Detector Masters Thesis Federal University of Minas Gerais, 2016. Resumo | Links | BibTeX | Tags: DeepEyes, DET, GigaFrames, Pedestrian Detection, VER+ @mastersthesis{Jordao:2016:MSc, title = {The Good, the Fast and the Better Pedestrian Detector}, author = {Artur Jordao}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/03/dissertation_2016_ArturJordao.pdf}, year = {2016}, date = {2016-06-24}, school = {Federal University of Minas Gerais}, abstract = {Pedestrian detection is a well-known problem in Computer Vision, mostly because of its direct applications in surveillance, transit safety and robotics. In the past decade, several efforts have been performed to improve the detection in terms of accuracy, speed and feature enhancement. In this work, we propose and analyze techniques focusing on these points. First, we develop an accurate oblique random forest (oRF) associated with Partial Least Squares (PLS). The method utilizes the PLS to find a decision surface, at each node of a decision tree, that better splits the samples presented to it, based on some purity criterion. To measure the advantages provided by PLS on the oRF, we compare the proposed method with the oRF based on SVM. Second, we evaluate and compare filtering approaches to reduce the search space and keep only potential regions of interest to be presented to detectors, speeding up the detection process. Experimental results demonstrate that the evaluated filters are able to discard a large number of detection windows without compromising the accuracy. Finally, we propose a novel approach to extract powerful features regarding the scene. The method combines results of distinct pedestrian detectors by reinforcing the human hypothesis, whereas suppressing a significant number of false positives due to the lack of spatial consensus when multiple detectors are considered. Our proposed approach, referred to as Spatial Consensus (SC), outperforms all previously published state-of-the-art pedestrian detection methods.}, keywords = {DeepEyes, DET, GigaFrames, Pedestrian Detection, VER+}, pubstate = {published}, tppubtype = {mastersthesis} } Pedestrian detection is a well-known problem in Computer Vision, mostly because of its direct applications in surveillance, transit safety and robotics. In the past decade, several efforts have been performed to improve the detection in terms of accuracy, speed and feature enhancement. In this work, we propose and analyze techniques focusing on these points. First, we develop an accurate oblique random forest (oRF) associated with Partial Least Squares (PLS). The method utilizes the PLS to find a decision surface, at each node of a decision tree, that better splits the samples presented to it, based on some purity criterion. To measure the advantages provided by PLS on the oRF, we compare the proposed method with the oRF based on SVM. Second, we evaluate and compare filtering approaches to reduce the search space and keep only potential regions of interest to be presented to detectors, speeding up the detection process. Experimental results demonstrate that the evaluated filters are able to discard a large number of detection windows without compromising the accuracy. Finally, we propose a novel approach to extract powerful features regarding the scene. The method combines results of distinct pedestrian detectors by reinforcing the human hypothesis, whereas suppressing a significant number of false positives due to the lack of spatial consensus when multiple detectors are considered. Our proposed approach, referred to as Spatial Consensus (SC), outperforms all previously published state-of-the-art pedestrian detection methods. |
Cristianne Rodrigues Santos Dutra Técnicas Otimizadas para Reidentificaçâo de Pessoas Masters Thesis Federal University of Minas Gerais, 2016. Links | BibTeX | Tags: DeepEyes, GigaFrames, Person Re-Identification, VER+ @mastersthesis{Dutra:2016:MSc, title = {Técnicas Otimizadas para Reidentificaçâo de Pessoas}, author = {Cristianne Rodrigues Santos Dutra}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/02/thesis_2016_Cristianne.pdf}, year = {2016}, date = {2016-01-01}, school = {Federal University of Minas Gerais}, keywords = {DeepEyes, GigaFrames, Person Re-Identification, VER+}, pubstate = {published}, tppubtype = {mastersthesis} } |
2015 |
Cassio Elias Santos dos Junior Partial Least Squares for Face Hashing Masters Thesis Federal University of Minas Gerais, 2015. Resumo | Links | BibTeX | Tags: DeepEyes, Face Identification, Face Recognition, GigaFrames, Indexing Structure, Local Sensitive Hashing, Partial Least Squares, VER+ @mastersthesis{Santos:2015:MSc, title = {Partial Least Squares for Face Hashing}, author = {Cassio Elias Santos dos Junior}, url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/02/dissertation_2015_Cassio.pdf}, year = {2015}, date = {2015-08-24}, school = {Federal University of Minas Gerais}, abstract = {Face identification is an important research topic due to areas such as its application to surveillance, forensics and human-computer interaction. In the past few years, a myriad of methods for face identification has been proposed in the literature, with just a few among them focusing on scalability. In this work, we propose a simple but efficient approach for scalable face identification based on partial least squares (PLS) and random independent hash functions inspired by locality-sensitive hashing (LSH), resulting in the PLS for hashing (PLSH) approach. The original PLSH approach is further extended using feature selection to reduce the computational cost to evaluate the PLS-based hash functions, resulting in the state-of-the-art extended PLSH approach (ePLSH). The proposed approach is evaluated in the dataset FERET and in the dataset FRGCv1. The results show significant reduction in the number of subjects evaluated in the face identification (reduced to 0.3% of the gallery), providing averaged speedups up to 233 times compared to evaluating all subjects in the face gallery and 58 times compared to previous works in the literature.}, keywords = {DeepEyes, Face Identification, Face Recognition, GigaFrames, Indexing Structure, Local Sensitive Hashing, Partial Least Squares, VER+}, pubstate = {published}, tppubtype = {mastersthesis} } Face identification is an important research topic due to areas such as its application to surveillance, forensics and human-computer interaction. In the past few years, a myriad of methods for face identification has been proposed in the literature, with just a few among them focusing on scalability. In this work, we propose a simple but efficient approach for scalable face identification based on partial least squares (PLS) and random independent hash functions inspired by locality-sensitive hashing (LSH), resulting in the PLS for hashing (PLSH) approach. The original PLSH approach is further extended using feature selection to reduce the computational cost to evaluate the PLS-based hash functions, resulting in the state-of-the-art extended PLSH approach (ePLSH). The proposed approach is evaluated in the dataset FERET and in the dataset FRGCv1. The results show significant reduction in the number of subjects evaluated in the face identification (reduced to 0.3% of the gallery), providing averaged speedups up to 233 times compared to evaluating all subjects in the face gallery and 58 times compared to previous works in the literature. |