Evaluating performance of feature extraction methods for practical 3D imaging systems

Dwarakanath, Deepak; Eichhorn, Alexander; Halvorsen, Pål; Griwodz, Carsten

doi:10.1145/2425836.2425887

Cited by 9 publications

(8 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Although many variants and alternatives have been developed, such as its approximation SURF [15] and the binary descriptor BRIEF [16], investigations demonstrate that SIFT is still more robust to viewpoint changes and common image disturbances than both BRIEF and SURF [17]. ORB [18], which is a combination of the FAST detector [19] and the BRIEF binary descriptor, is a good choice for real-time applications, but several evaluations state that it cannot reach the repeatability and discriminative properties of SIFT [20][21][22][23]. KAZE [24] is a new development and succeeds especially in the presence of deformable objects.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Automatic UAV Image Geo-Registration by Matching UAV Images to Georeferenced Image Data

et al. 2017

View full text Add to dashboard Cite

Recent years have witnessed the fast development of UAVs (unmanned aerial vehicles).As an alternative to traditional image acquisition methods, UAVs bridge the gap between terrestrial and airborne photogrammetry and enable flexible acquisition of high resolution images. However, the georeferencing accuracy of UAVs is still limited by the low-performance on-board GNSS and INS. This paper investigates automatic geo-registration of an individual UAV image or UAV image blocks by matching the UAV image(s) with a previously taken georeferenced image, such as an individual aerial or satellite image with a height map attached or an aerial orthophoto with a DSM (digital surface model) attached. As the biggest challenge for matching UAV and aerial images is in the large differences in scale and rotation, we propose a novel feature matching method for nadir or slightly tilted images. The method is comprised of a dense feature detection scheme, a one-to-many matching strategy and a global geometric verification scheme. The proposed method is able to find thousands of valid matches in cases where SIFT and ASIFT fail. Those matches can be used to geo-register the whole UAV image block towards the reference image data. When the reference images offer high georeferencing accuracy, the UAV images can also be geolocalized in a global coordinate system. A series of experiments involving different scenarios was conducted to validate the proposed method. The results demonstrate that our approach achieves not only decimeter-level registration accuracy, but also comparable global accuracy as the reference images.

show abstract

Section: Related Workmentioning

confidence: 99%

“…Among the state-of-the-art matching algorithms, SIFT has been proven to be scale and rotation invariant and to outperform other local descriptors in various evaluations [20][21][22][23]. Besides, the ratio test proposed by Lowe [7] is widely applied to discard mismatches.…”

Section: Siftmentioning

confidence: 99%

Automatic UAV Image Geo-Registration by Matching UAV Images to Georeferenced Image Data

et al. 2017

View full text Add to dashboard Cite

show abstract

“…Avaliações de desempenho têm cada vez mais importância em visão computacional [25], haja vista a quantidade de trabalhos nesta área. Dentre esses, muitos trabalhos surgiram com objetivo de avaliar algoritmos de correspondência de pontos em imagens, por meio de diferentes estratégias e métricas avaliativas [25].…”

Section: Trabalhos Relacionadosunclassified

“…Dentre esses, muitos trabalhos surgiram com objetivo de avaliar algoritmos de correspondência de pontos em imagens, por meio de diferentes estratégias e métricas avaliativas [25].…”

Section: Trabalhos Relacionadosunclassified

Análise comparativa de detectores e descritores de características locais em imagens no âmbito do problema de autocalibração de câmeras

Brito

Pádua

Lopes

et al. 2016

RBCA

View full text Add to dashboard Cite

Resumo: Este trabalho apresenta uma análise comparativa de diferentes métodos do estado da arte para detecção e descrição de características locais em imagens, com o objetivo de solucionar de forma robusta e eficiente o problema de autocalibração de câmeras. Para atingir esse objetivo, é essencial a utilização de métodos detectores e descritores eficazes, uma vez que a correspondência robusta de características em um conjunto de imagens sucessivas sujeitas a uma ampla variedade de distorções afins e mudanças no ponto de vista 3D da cena, é crucial para a exatidão dos cálculos dos parâmetros da câmera. Muito embora diversos detectores e descritores têm sido propostos na literatura, seus impactos no processo de autocalibração de câmeras não foram ainda devidamente estudados. Nesse trabalho de análise comparativa, utilizam-se como critérios de qualidade da autocalibração os erros: epipolar, de reprojeção e reconstrução, bem como os tempos de execução dos métodos. Os resultados experimentais demonstram que detectores e descritores binários de características (ORB, BRISK e FREAK) e de ponto flutuante (SIFT e SURF) apresentam erros de reprojeção e reconstrução equivalentes. Considerando-se, porém, o menor custo computacional dos métodos binários, recomenda-se, fortemente, o uso destes em soluções de problemas de autocalibração de câmeras. Palavras-chave: Autocalibração de Câmeras, Detectores e Descritores Locais, Correspondência de Pontos em Imagens.Abstract: This work presents a comparative analysis of different state-of-art local feature detectors and descriptors aiming to accurately and efficiently solve the camera self-calibration problem. To achieve this goal, it is essential the usage of effective image feature detectors and descriptors, since the robust matching of features in a set of successive images subjected to a substantial range of affine distorsion and changes in 3D viewpoint is crucial to the accurate computation of the camera parameters. Even though a lot of detectors and descriptors have been proposed in the literature, their impacts in the camera self-calibration task have not yet been carefully studied. Our comparative analysis uses as criteria the estimated epipolar, reprojection and reconstruction errors, as well as the running times of the methods. Our experiments show that binary (ORB, BRISK and FREAK) and floating-point (SIFT and SURF) feature detectors and descriptors present equivalent reprojection and reconstruction errors. Therefore,considering the smaller computational cost of binary methods, their usage is strongly recommended in camera self-calibration problems.

show abstract

“…It measures the ability to obtain sufficient feature point correspondences in the images [23] .After getting and observing all the features as shown in figure 7 a conclusion tells that the technique so use has a good detect-ability.…”

mentioning

confidence: 98%

Accelerating CBIR System using Graphics Processing Unit in OPEN CV Environment

Kaur¹,

Jindal²

2015

IJCA

View full text Add to dashboard Cite

Content Based Image Retrieval is a system for retrieving the images from a dataset of images. The CBIR has several stages including feature extraction at the initial level and then performing the similarity matching. Later on based upon the similarity matching results the required images can be extracted. Images are a complex data to handle as they are composed of matrices and vectors of data and also due to multi thread execution of algorithms, programmability and low cost, image processing becomes an appropriate field of achieving parallelism. To parallelize the phases of CBIR a special hardware is used known as Graphics Processing Unit. The objectives of the published work is to develop a Content Based Image Retrieval system and test it on benchmark WANG database using suitable and efficient techniques so as to achieve efficient results. As images are complex and there is an approach of achieving parallelism while dealing with images, so after implementing CBIR, the feature extraction phase has been parallelized and the tendency of GPU processor is exploited to obtain the speed up. An accuracy of range between 22 to 67% for individual categories of the databasewas achieved. Finally on comparing the execution times of serial implementation and parallel implementation a considerable speed up of 15 % on an average was obtained.

show abstract

Evaluating performance of feature extraction methods for practical 3D imaging systems

Cited by 9 publications

References 18 publications

Automatic UAV Image Geo-Registration by Matching UAV Images to Georeferenced Image Data

Automatic UAV Image Geo-Registration by Matching UAV Images to Georeferenced Image Data

Análise comparativa de detectores e descritores de características locais em imagens no âmbito do problema de autocalibração de câmeras

Accelerating CBIR System using Graphics Processing Unit in OPEN CV Environment

Contact Info

Product

Resources

About