A Deformabled Object Bank Model for Multimedia Event Detection

2012.09- Present Joined the School of Computer Science at Carnegie Mellon University, worked in the Informedia Group as a Visiting Scholar, and focused research mainly on Multimedia Retrieval & Multimedia Event Detection.
Description—In order to detect efficiently lots events in Surveillance video, we use a Deformabled Object Bank approach, which is based on Deformable Part Model(Felzenszwalb, CVPR 2008) and Object Bank Model(Li-Jia Li,NIPS 2010), to train and evaluate 1000 related object models by using the Image-Net and the Pascal VOC dataset.

Study on system modeling method and high performance computing method for regulation mechanism of tumor growth

2010.09-2011.11 Participated in a project entitled Study on system modeling method and high performance computing method for regulation mechanism of tumor growth, which was supported by National Natural Science Foundation of China with Grant No. 61070092.
Abstract—Cancer, commonly known as malignant tumors, because of its high-aggressive, highly lethal and can not be cured, and become one of the main diseases to threat to human life and health. To this end, each countries has put a lot of manpower, material and financial resources in the field of cancer research. Because the tumor is a highly complex self-organizing dynamic system, if tumor growth can be established by studying the mathematical model, depicts the whole process of tumor growth and found that the internal mechanism of tumor growth, it will bring new hope to the treatment and prevention tumor diseases. Currently, the popular cell tumor growth model is a Potts model. But potts model has strictly requirements on tumor geometry and expensive computation. Finite element in this respect have advantage over other models, for the above reason ,we use the finite element method (FEM) for simulation and accomplished single cell and multi-cell growth simulation by introducing parallel computing method.

Study on modular & digital design for filling equipment

2008.12-2010.07 Participated in a project entitled Study on modular & digital design for filling equipment, which was supported by National Spark Program of China with Grant No.2008GA780039.Introduced High Level Architecture to production line design. Put forward a service-oriented and HLA-based simulation model of juice production line (SHLA-SJPL).
Abstract—According to the deficiency that existed in traditional design mode of production line and the fact that the federate ambassador callback between the LRC(local RTI) and the CRC(central RTI) can hardly pass through firewalls in grid when using HLA (High Level Architecture), this paper puts forward a service-oriented and HLA-based simulation model of juice production line(SHLA-SJPL) and gives a model design as well as realization based on modeling software Solidworks, scene simulation software EON, MAK RTI and VC++. Subsequent simulation results show that the model can shorten effectively the design period of juice production line as well as improve greatly the design quality and is of great reference value for the construction and development of HLA-based simulation system in other fields.

Study on precise location for vision related fruit picking mechanism based on intelligent design

2007.09-2008.04 Participated in a project entitled Study on precise location for vision related fruit picking mechanism based on intelligent design, which was supported by National Natural Science Foundation of China with Grant No. 50775079. Studied mainly for multi-source images based uncertain information fusion problem, and proposed a fusion approach of trinocular vision in wavelet domain based on fuzzy reasoning weight, which improving greatly the adaptive recognition abilities of picking robot in complex environment.
Abstract—Aiming to uncertain problem in the process of picking manipulator location and behavior decision-making in complex environment, an approach for picking manipulator location and behavior simulation integrating binocular stereo vision together virtual simulation technical was discussed. Firstly, the imaging of picking target and environment were gained by binocular stereo vision, secondly, the result data of picking target and environment after the imaging being handled were mapped into virtual environment, picking manipulator behavior decision-making and motion path planning were implemented by software simulation technical, picking manipulator accurate location was realized in virtual environment. Finally, the simulation motion data in virtual environment were transmitted to control motors at manipulator joint by USB interface, control motors drive manipulator moving automatically to the picking target position. The experimental platform was consisted of CCD stereo vision and simulation software, a location simulation system was developed by VC++ and EON SDK.