Feb 27, 2007: Person Verification by 3D Surface Alignment
Filed in: Colloquium
Dr. Dirk Colbry, Michigan State University
Automatic person verification is desirable for e-commerce, travel, and other security concerns. Using the human face as a biometric has many benefits, and the 3D face scan appears to have advantages over 2D images. The 3DID system has been developed at MSU with support from the Michigan Economic Development Corporation, and will be commercially viable once the cost constraints are met. The current prototype uses the Minolta Vivid 910 scanner to acquire a 320 x 240 image of "pixels & points" [R, G, B, x, y, z]. This image can be compared to a model of the face stored on a computer or smart card device, such as an ID or credit card. The 3DID system matches the input scan to the stored model using automatic anchor point detection for coarse alignment, followed by a fast trimmed Iterative Closest Point (ICP) algorithm to compute the best fine alignment of the two face surfaces. Color is not yet used. The algorithm is fast (less than 2 seconds on a 3.2 GHz P4) and can tolerate translations of up to 15 degrees of roll and pitch, and up to 30 degrees of yaw, relative to a frontal, neutral pose of the face. The system has been successfully field tested in a number of locations, including the Lansing Airport, and will be demonstrated live during the presentation.
Dr. Dirk Colbry earned a B.M.E. from the Georgia Institute of Technology, a M.S.E. (Computer Science and Engineering) from the University of Michigan, and a Ph.D. (Computer Science) from Michigan State University. At MSU, he also completed the interdepartmental graduate specialization in Cognitive Science and was awarded an NSF IGERT Associate Fellowship. Currently, Dr. Colbry is a visiting assistant professor in the Cognitive Science program at Michigan State. Dr. Colbry's industry experience includes engineering work for Delta Airlines and FANUC Robotics North America, as well consulting on a number of industry projects related to his current research in computer vision and pattern recognition. His research experience includes projects in artificial intelligence and intelligent planning systems (specifically, the use of Bayesian networks to develop planning systems for cognitive robotic aids for the elderly), as well as work in machine vision and pattern recognition. His dissertation, entitled "Human Face Verification by Robust 3D Surface Alignment," described his research using 3D scanning technology in security applications. In addition to extending his dissertation research, Dr. Colbry is collaborating with industrial and academic partners to apply pattern recognition and image processing research to problems in psychology, zoology, cognitive science, telecommunications, medicine and mechanical engineering.
Abstract
Automatic person verification is desirable for e-commerce, travel, and other security concerns. Using the human face as a biometric has many benefits, and the 3D face scan appears to have advantages over 2D images. The 3DID system has been developed at MSU with support from the Michigan Economic Development Corporation, and will be commercially viable once the cost constraints are met. The current prototype uses the Minolta Vivid 910 scanner to acquire a 320 x 240 image of "pixels & points" [R, G, B, x, y, z]. This image can be compared to a model of the face stored on a computer or smart card device, such as an ID or credit card. The 3DID system matches the input scan to the stored model using automatic anchor point detection for coarse alignment, followed by a fast trimmed Iterative Closest Point (ICP) algorithm to compute the best fine alignment of the two face surfaces. Color is not yet used. The algorithm is fast (less than 2 seconds on a 3.2 GHz P4) and can tolerate translations of up to 15 degrees of roll and pitch, and up to 30 degrees of yaw, relative to a frontal, neutral pose of the face. The system has been successfully field tested in a number of locations, including the Lansing Airport, and will be demonstrated live during the presentation.
Bio
Dr. Dirk Colbry earned a B.M.E. from the Georgia Institute of Technology, a M.S.E. (Computer Science and Engineering) from the University of Michigan, and a Ph.D. (Computer Science) from Michigan State University. At MSU, he also completed the interdepartmental graduate specialization in Cognitive Science and was awarded an NSF IGERT Associate Fellowship. Currently, Dr. Colbry is a visiting assistant professor in the Cognitive Science program at Michigan State. Dr. Colbry's industry experience includes engineering work for Delta Airlines and FANUC Robotics North America, as well consulting on a number of industry projects related to his current research in computer vision and pattern recognition. His research experience includes projects in artificial intelligence and intelligent planning systems (specifically, the use of Bayesian networks to develop planning systems for cognitive robotic aids for the elderly), as well as work in machine vision and pattern recognition. His dissertation, entitled "Human Face Verification by Robust 3D Surface Alignment," described his research using 3D scanning technology in security applications. In addition to extending his dissertation research, Dr. Colbry is collaborating with industrial and academic partners to apply pattern recognition and image processing research to problems in psychology, zoology, cognitive science, telecommunications, medicine and mechanical engineering.