老校训：仰天地之正气，法古今之完人。
同济老校歌，铿锵有力
校歌歌词
好一片中华大地
不振兴工艺
真可惜 真可惜
同有耳目
同有手足
同有心思才力
不作工负了好教育
勤劳 诚毅
提携我中华国民
同舟共济 同舟共济
振兴工艺
好一片中华大地
不健康身体
真可惜 真可惜
同有心腹 同有肌肉
同有起居饮食
不学医负了好教育
慈爱 仁义
扶持我中华国民
同舟共济 同舟共济
健康身体
好一片中华大地
不格物穷理
真可惜 真可惜
同有头脑 同有智慧
同有星辰空气
不学理负了好教育
明彻 清晰
训练我中华国民
同舟共济 同舟共济
格物穷理
（注：此校歌创作于1927年）
Tuesday, March 20, 2007
Accuracy and Geometric Decomposition
Lack of automatic support for data collection and interpretation makes the laser scanning based bridge geometric data acquisition process prohibitive to ordinary bridge inspectors. First, even though from the technical data of a laser scanner bridge inspectors can know how accurate each point can be positioned, what are more useful for them are the accuracy of geometric features which can be extracted from dense point cloud such as how much uncertainty lies in the normal values of an extracted plane. Evaluation of those geometric features and extraction of high level uncertainty information require bridge inspectors to know how to do manual uncertainty analysis. In addition, that process is too complicated for on-site data evaluation to get a quick feedback about the data quality. Second, bridge inspectors need to manually analyze the relationship between the accuracy of single geometric features and the value generated based on them. Interestingly, accuracy constraints for a group of geometric features can have multiple solutions to achieve the same accuracy level for their corresponding inspection goal. So if bridge inspector can get a tool to show them the trade-off between the accuracy constraints to achieve specific accuracy requirement for an inspection goal so that they can have more alternatives on site and get more flexibility on site. Third, bridge inspectors need to interpret accuracy into a series of surveying strategies such as locate the scanners within
Sunday, March 11, 2007
Today's work
1. Finish the executive summary of the PhD Proposal
2. Search papers related to my research and read the paper from Carl Haas and a paper from Burcu (About the impact of new data collection technology on data collection and transfer process for timely construction management, Journal of Construction Engineering and Management)
3. Find a paper from Martin Fisher: Formalizing Product Model Transformations (This paper might be important for me to understand how to extend IFC bridge information model to support inspection goal decomposition since it transforms design centered model into a construction activity centered model through three mechanisms introduced by the authors, I can transform the IFC bridge model to a surveying activity centered model so that bridge inspection goal will be transformed into a series of measurement activities to generate inspection goal value: such as the value of the under clearance of a bridge, we can decompose it into measurement of the bottom surface of the super structure and the measurement of the top surface of the road underneath the bridge).
Idea:
From the bridge function(cross river or valley or another highway, we can navigate to the road underneath the bridge, and then from the attribute of the bridge such as how big vehicles are allowed to run on it we can simulate the process that the biggest allowed truck passing under the bridge and to see whether the bridge's under clearance is satisfactory). This reasoning process is transparent, because it is not just specify a limit value for the under clearance, it shows the design intent behind the under clearance limitation values.
2. Search papers related to my research and read the paper from Carl Haas and a paper from Burcu (About the impact of new data collection technology on data collection and transfer process for timely construction management, Journal of Construction Engineering and Management)
3. Find a paper from Martin Fisher: Formalizing Product Model Transformations (This paper might be important for me to understand how to extend IFC bridge information model to support inspection goal decomposition since it transforms design centered model into a construction activity centered model through three mechanisms introduced by the authors, I can transform the IFC bridge model to a surveying activity centered model so that bridge inspection goal will be transformed into a series of measurement activities to generate inspection goal value: such as the value of the under clearance of a bridge, we can decompose it into measurement of the bottom surface of the super structure and the measurement of the top surface of the road underneath the bridge).
Idea:
From the bridge function(cross river or valley or another highway, we can navigate to the road underneath the bridge, and then from the attribute of the bridge such as how big vehicles are allowed to run on it we can simulate the process that the biggest allowed truck passing under the bridge and to see whether the bridge's under clearance is satisfactory). This reasoning process is transparent, because it is not just specify a limit value for the under clearance, it shows the design intent behind the under clearance limitation values.
Friday, March 9, 2007
Flash LADAR
Flash LARDAR is a new sensor. We can call it 3D camera since it can simultaneously collect thousands of 3D points like a camera can simultaneously record color and reflectivity information for thousands of pixels.
Professor Carl Haas at University of Toronto (University of Texas Austin previously) are trying to use this new technology for spatial management on the construction site. This new technique can collect accurate 3D information directly and avoid the limitation of digital camera which requires complicated model to reconstruct depth information from multiple photos.
Professor Haas is also trying to merge the information from CAD model with 3D point clouds to enhance the speed and accuracy rate of traditional computer vision technique since in CAD model there is prerequisite knowledge about the 3D scene. But how to formally represent those knowledge in 3D model so that computer can automatically interpret 3D semantic information from 3D CAD model is a problem. The representation in IFC (a paper of Professor Grundig at UT Berlin talks about this from the perspective of a surveyor) and the semantic web technology might be of use to formally capture semantic information from 3D CAD model so that we can automatically extract knowledge from 3D model and help computer vision algorithms to interpret 3D point clouds which only contains xyz value and generate intelligent point clouds.
Good future is coming:-)
Professor Carl Haas at University of Toronto (University of Texas Austin previously) are trying to use this new technology for spatial management on the construction site. This new technique can collect accurate 3D information directly and avoid the limitation of digital camera which requires complicated model to reconstruct depth information from multiple photos.
Professor Haas is also trying to merge the information from CAD model with 3D point clouds to enhance the speed and accuracy rate of traditional computer vision technique since in CAD model there is prerequisite knowledge about the 3D scene. But how to formally represent those knowledge in 3D model so that computer can automatically interpret 3D semantic information from 3D CAD model is a problem. The representation in IFC (a paper of Professor Grundig at UT Berlin talks about this from the perspective of a surveyor) and the semantic web technology might be of use to formally capture semantic information from 3D CAD model so that we can automatically extract knowledge from 3D model and help computer vision algorithms to interpret 3D point clouds which only contains xyz value and generate intelligent point clouds.
Good future is coming:-)
Tuesday, March 6, 2007
ICES Presentation of Pingbo
My presentation slides about geometric defect detection of Bombardier track system is available from the following link:
http://www.ices.cmu.edu/censcir/participate/collaboration_event_archives.html
http://www.ices.cmu.edu/censcir/participate/collaboration_event_archives.html
Specification Model and Inspection Planning for Automatic Inspection Goal Decomposition
The researches of Frank Boukamp and Chris Gordon can be combined together and serve as the foundation of my research, which is about geometric inspection goal decomposition.
Frank is modeling the specification so that the knowledge in construction site inspection specifications can be represented in a formal way so that computer can interpret the knowledge from the specification model automatically and use the knowledge to detect defects of a structure. Here, defect means the part of a structure violating the design specification.
From Frank's model, Chris can generate a list of inspection task for a specific project based on the context of the site, the preference of the user and the project schedule.
I can use Frank's model to generate a list of bridge inspection goals, then I match those project specific goal with a generic inspection goal defined by myself. The generic inspection goal will separate domain abstrct knowledge from specific case. For example, I want to know the under clearance of a specific bridge, but the knowledge it self does not have to be related with a specific bridge, we can just say we want to know the size of a empty space. Then we decompose the generic geometric inspection goal into a number of generic targets. Then I matching those generic targets back to the specific context and make those generic targets to be specific. For example, I decompose "measuring vertical space size" into "measure bottom surface of the solid object over the space" and "measure the top surface of the solid object below the space". The using the project specific information I will know that the solid object below the space is the top the highway below the bridge (which can be extracted from GIS system or bridge management system), and the object above the space is the super structure of the bridge. Then I extract these specific information and get the bottom surface of the above object(super structure) and the top surface of the below object (highway). Finally I calculate the vertical distance between these two surfaces and get the under clearance of a specific bridge.
From specific inspetion goal to generic inspection goal, from inspection goal to generic measurement target, then using the specific project information the generate specific measurement targets, the whole project is inspection goal decomposition and assembly which provide a automatic process for bridge inspection gaol acquisition.
Frank is modeling the specification so that the knowledge in construction site inspection specifications can be represented in a formal way so that computer can interpret the knowledge from the specification model automatically and use the knowledge to detect defects of a structure. Here, defect means the part of a structure violating the design specification.
From Frank's model, Chris can generate a list of inspection task for a specific project based on the context of the site, the preference of the user and the project schedule.
I can use Frank's model to generate a list of bridge inspection goals, then I match those project specific goal with a generic inspection goal defined by myself. The generic inspection goal will separate domain abstrct knowledge from specific case. For example, I want to know the under clearance of a specific bridge, but the knowledge it self does not have to be related with a specific bridge, we can just say we want to know the size of a empty space. Then we decompose the generic geometric inspection goal into a number of generic targets. Then I matching those generic targets back to the specific context and make those generic targets to be specific. For example, I decompose "measuring vertical space size" into "measure bottom surface of the solid object over the space" and "measure the top surface of the solid object below the space". The using the project specific information I will know that the solid object below the space is the top the highway below the bridge (which can be extracted from GIS system or bridge management system), and the object above the space is the super structure of the bridge. Then I extract these specific information and get the bottom surface of the above object(super structure) and the top surface of the below object (highway). Finally I calculate the vertical distance between these two surfaces and get the under clearance of a specific bridge.
From specific inspetion goal to generic inspection goal, from inspection goal to generic measurement target, then using the specific project information the generate specific measurement targets, the whole project is inspection goal decomposition and assembly which provide a automatic process for bridge inspection gaol acquisition.
Thursday, March 1, 2007
Intelligent Point cloud for 3D data registration
If we can not recognize 3D objects from 3D point clouds, those point clouds are just a bunch of xyz coordinates. Today I found that for complicated environments like inside the vision group laboratory, even though there is only minor changes of the environment (walking of a student, movement of a small board), two scans can not be registered together accurately.
That's too bad, because we can never expect construction site will be clean and static. If we can not automatically recognize object from point clouds and separate noisy information from useful information, we can not register the 3D point clouds using traditional algorithms. Another thing is that if we want to detect the environment change, we can not directly register two model and get the differences between them, we must segment the point clouds into patches and do "patch-based matching" between two 3D models and find a transformation which can figure out "outlier patches", which is the actual differences between two scenarios.
Anyway, if we know in advance which point belongs to which object, computer and operate on a "object" level and utilize domain knowledge to reason about the scene and detect changes or find a data registration results making more sense(this is a wall, it will not change, that is a table, it could be moved), so we will avoid the problem caused by non-semantic approach: all points are viewed as the same thing(a point, nothing else!), and the algorithm simply trying to minimize the distance between points so that small changes in the scenario will influence the whole registration results and result in failure of 3D registration.
That's too bad, because we can never expect construction site will be clean and static. If we can not automatically recognize object from point clouds and separate noisy information from useful information, we can not register the 3D point clouds using traditional algorithms. Another thing is that if we want to detect the environment change, we can not directly register two model and get the differences between them, we must segment the point clouds into patches and do "patch-based matching" between two 3D models and find a transformation which can figure out "outlier patches", which is the actual differences between two scenarios.
Anyway, if we know in advance which point belongs to which object, computer and operate on a "object" level and utilize domain knowledge to reason about the scene and detect changes or find a data registration results making more sense(this is a wall, it will not change, that is a table, it could be moved), so we will avoid the problem caused by non-semantic approach: all points are viewed as the same thing(a point, nothing else!), and the algorithm simply trying to minimize the distance between points so that small changes in the scenario will influence the whole registration results and result in failure of 3D registration.
Subscribe to:
Posts (Atom)
Posts
