Yeqing Li - Projects

Large-scale Image Search
In this project, we aims to building an large-scale image search system. There are two key challenges of the image search problem: (1) We need to find an effective way to describe each images (i.e. features); (2) We need to find a efficient way to perform the search. For the second challenge, we develop a novel image hash algorithm that transform the high-dimensional floating-point features to short binary codes. The resulting binary codes are more compact to store and efficient to search.

Large-scale Multi-view Unsupervised Learning
In real world applications, data can be describe in many different features, such as different languages for text, different visual descriptor for images, different measures of genes. Different features usually describe different aspects of data and compensate to each other. In this project, we develop new algorithm for integrating heterogeneous features for better performance. We aims to address the multi-view learning problem in large-scale data.

Computer Vision Aided Geographic Information Mining
In this project, we develop algorithms to inference detailed geographic information of highways, such as road signs, lanes and marks, given vehicle-sensed data. The raw data is collected by vehicle-mounted devices, e.g. camera and GPS, on many vehicles. Then, numerical features are extracted by running computer vision algorithm on the raw data. We develop algorithms to inference the accurate positions of the geographic objects. The final results are given in geographic coordinates, which have reached to an accuracy of just meters.

Non-Intrusive Load Monitoring (NILM)
In this project, we study the problem of NILM, which aims to inference the energy consumption of each individual appliance from the whole home energy consumption. We build models to capture the correlation between energy consumption of difference appliances and to utilize the external factors to improve the accuracy of the inference.

Real-time Face Tracking
In this project, we build a real-time face tracking system with camera by using Active Shape Model (ASM) and Constraint Local Model (CLM). We also integrate facial expression analysis and head gesture detection components into the system for human behavior study.

Online Object Tracking
In this project, we build a general object tracking system based on online learning. This system has been applied to very challenging video sequences of surgical tool tracking and achieved very promising results.