Eduardo R. Corral-Soto, R. Tal, Langyue Wang, R. Persad, Luo Chao, C. Solomon, Bob Hou, G. Sohn, J. Elder
The 3DTown project is focused on the development of a distributed system for sensing, interpreting and visualizing the real-time dynamics of urban life within the 3D context of a city. At the heart of this technology lies a core of algorithms that automatically integrate 3D urban models with data from pan/tilt video cameras, environmental sensors and other real-time information sources. A key challenge is the "three-dimensionalization" of pedestrians and vehicles tracked in 2D camera video, which requires automatic real-time computation of camera pose relative to the 3D urban environment. In this paper we report preliminary results from a prototype system we call 3DTown, which is composed of discrete modules connected through pre-determined communication protocols. Currently, these modules consist of: 1) A 3D modeling module that allows for the efficient reconstruction of building models and integration with indoor architectural plans, 2) A GeoWeb server that indexes a 3D urban database to render perspective views of both outdoor and indoor environments from any requested vantage, 3) Sensor modules that receive and distribute real-time data, 4) Tracking modules that detect and track pedestrians and vehicles in urban spaces and access highways, 5) Camera pose modules that automatically estimate camera pose relative to the urban environment, 6) Three-dimensionalization modules that receive information from the GeoWeb server, tracking and camera pose modules in order to back-project image tracks to geolocate pedestrians and vehicles within the 3D model, 7) An animation module that represents geo-located dynamic agents as sprites, and 8) A web-based visualization module that allows a user to explore the resulting dynamic 3D visualization in a number of interesting ways. To demonstrate our system we have used a blend of automatic and semi-automatic methods to construct a rich and accurate 3D model of a university campus, including both outdoor and indoor detail. The demonstration allows web-based 3D visualization of recorded patterns of pedestrian and vehicle traffic on streets and highways, estimations of vehicle speed, and real-time (live) visualization of pedestrian traffic and temperature data at a particular test site. Having demonstrated the system for hundreds of people, we report our informal observations on the user reaction, potential application areas and on the main challenges that must be addressed to bring the system closer to deployment.
{"title":"3D Town: The Automatic Urban Awareness Project","authors":"Eduardo R. Corral-Soto, R. Tal, Langyue Wang, R. Persad, Luo Chao, C. Solomon, Bob Hou, G. Sohn, J. Elder","doi":"10.1109/CRV.2012.64","DOIUrl":"https://doi.org/10.1109/CRV.2012.64","url":null,"abstract":"The 3DTown project is focused on the development of a distributed system for sensing, interpreting and visualizing the real-time dynamics of urban life within the 3D context of a city. At the heart of this technology lies a core of algorithms that automatically integrate 3D urban models with data from pan/tilt video cameras, environmental sensors and other real-time information sources. A key challenge is the \"three-dimensionalization\" of pedestrians and vehicles tracked in 2D camera video, which requires automatic real-time computation of camera pose relative to the 3D urban environment. In this paper we report preliminary results from a prototype system we call 3DTown, which is composed of discrete modules connected through pre-determined communication protocols. Currently, these modules consist of: 1) A 3D modeling module that allows for the efficient reconstruction of building models and integration with indoor architectural plans, 2) A GeoWeb server that indexes a 3D urban database to render perspective views of both outdoor and indoor environments from any requested vantage, 3) Sensor modules that receive and distribute real-time data, 4) Tracking modules that detect and track pedestrians and vehicles in urban spaces and access highways, 5) Camera pose modules that automatically estimate camera pose relative to the urban environment, 6) Three-dimensionalization modules that receive information from the GeoWeb server, tracking and camera pose modules in order to back-project image tracks to geolocate pedestrians and vehicles within the 3D model, 7) An animation module that represents geo-located dynamic agents as sprites, and 8) A web-based visualization module that allows a user to explore the resulting dynamic 3D visualization in a number of interesting ways. To demonstrate our system we have used a blend of automatic and semi-automatic methods to construct a rich and accurate 3D model of a university campus, including both outdoor and indoor detail. The demonstration allows web-based 3D visualization of recorded patterns of pedestrian and vehicle traffic on streets and highways, estimations of vehicle speed, and real-time (live) visualization of pedestrian traffic and temperature data at a particular test site. Having demonstrated the system for hundreds of people, we report our informal observations on the user reaction, potential application areas and on the main challenges that must be addressed to bring the system closer to deployment.","PeriodicalId":372951,"journal":{"name":"2012 Ninth Conference on Computer and Robot Vision","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2012-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125904196","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper proposes an extension to anisotropic diffusion filtering for a better preservation of semantically meaningful structures such as edges in an image in its smoothing/denoising process. The problem of separation of the gradients due to edges and the gradients due to noise is formulated as a nonlinearly separable classification problem. More specifically, the spatially-regularized image gradient is mapped to a higher dimensional Reproducing Kernel Hilbert Space (RKHS) in which the gradients of the edges from those of noise can be readily separated. This proper discrimination of edges prevents the filter from blurring the edges, while smoothing the image. Compared to the existing anisotropic filters, the proposed method improves the denoising and smoothing of an image on both synthetic and real images.
{"title":"Regularized Gradient Kernel Anisotropic Diffusion for Better Image Filtering","authors":"A. Shabani, J. Zelek, David A Clausi","doi":"10.1109/CRV.2012.57","DOIUrl":"https://doi.org/10.1109/CRV.2012.57","url":null,"abstract":"This paper proposes an extension to anisotropic diffusion filtering for a better preservation of semantically meaningful structures such as edges in an image in its smoothing/denoising process. The problem of separation of the gradients due to edges and the gradients due to noise is formulated as a nonlinearly separable classification problem. More specifically, the spatially-regularized image gradient is mapped to a higher dimensional Reproducing Kernel Hilbert Space (RKHS) in which the gradients of the edges from those of noise can be readily separated. This proper discrimination of edges prevents the filter from blurring the edges, while smoothing the image. Compared to the existing anisotropic filters, the proposed method improves the denoising and smoothing of an image on both synthetic and real images.","PeriodicalId":372951,"journal":{"name":"2012 Ninth Conference on Computer and Robot Vision","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2012-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126803603","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Matthias Rätsch, P. Quick, P. Huber, T. Frank, T. Vetter
In this paper, we introduce concepts to reduce the computational complexity of regression, which are successfully used for Support Vector Machines. To the best of our knowledge, we are the first to publish the use of a cascaded Reduced Set Vector approach for regression. The Wavelet-Approximated Reduced Vector Machine classifiers for face and facial feature point detection are extended to regression for efficient and robust head pose estimation. We use synthetic data, generated by the 3D Morph able Model, for optimal training sets and demonstrate results superior to state-of-the-art techniques. The new Wavelet Reduced Vector Regression shows similarly good results on natural data, gaining a reduction of the complexity by a factor of up to 560. The introduced Evolutionary Regression Tree uses coarse-to-fine loops of strongly reduced regression and classification up to most accurate complex machines. We demonstrate the Cascaded Condensation Tracking for head pose estimation for a large pose range up to ±90 degrees on videostreams.
{"title":"Wavelet Reduced Support Vector Regression for Efficient and Robust Head Pose Estimation","authors":"Matthias Rätsch, P. Quick, P. Huber, T. Frank, T. Vetter","doi":"10.1109/CRV.2012.41","DOIUrl":"https://doi.org/10.1109/CRV.2012.41","url":null,"abstract":"In this paper, we introduce concepts to reduce the computational complexity of regression, which are successfully used for Support Vector Machines. To the best of our knowledge, we are the first to publish the use of a cascaded Reduced Set Vector approach for regression. The Wavelet-Approximated Reduced Vector Machine classifiers for face and facial feature point detection are extended to regression for efficient and robust head pose estimation. We use synthetic data, generated by the 3D Morph able Model, for optimal training sets and demonstrate results superior to state-of-the-art techniques. The new Wavelet Reduced Vector Regression shows similarly good results on natural data, gaining a reduction of the complexity by a factor of up to 560. The introduced Evolutionary Regression Tree uses coarse-to-fine loops of strongly reduced regression and classification up to most accurate complex machines. We demonstrate the Cascaded Condensation Tracking for head pose estimation for a large pose range up to ±90 degrees on videostreams.","PeriodicalId":372951,"journal":{"name":"2012 Ninth Conference on Computer and Robot Vision","volume":"93 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2012-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114876516","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：