Research interests

My research interests are in computational vision, sensor and data fusion, environment modeling, perception for mobile robots and probabilistic methods.
During my PhD I worked on the problem of inferring complete range maps where only intensity images and very limited range data is available. The goal is to facilitate the building of 3D environment models by exploiting the fact that both video imaging and limited range sensing are ubiquitous readily-available technologies while complete volume scanning is prohibitive on most mobile platforms.

Research description

Range Synthesis for 3D Environment Modeling

This work was done during my PhD under the supervision of Gregory Dudek. We have developed a new statistical learning method to infer the geometric structures from images. Specifically, our method computes dense range maps of locations of the environment using only intensity images and very limited amount of range data as an input. Our goal is to facilitate the building of 3D environment models by exploiting the fact that both video imaging and limited range sensing are ubiquitous readily-available technologies while complete volume scanning is prohibitive on most mobile platforms. The main idea is to exploit the assumption that intensity and range data are correlated, albeit in potentially complicated ways, but exhibiting useful structure. The scientific issue is to represent this correlation such that it can be used to recover range data where missing. Markov Random Fields are used as a model to capture the local statistics of the intensity and range.

Please go to my list of publications for more detail on this subject.

Image Enhancement for Underwater Images

For many inspection and observation tasks, high quality image data is desirable. We have sucessfully applied our learning based Markov random field model to image enhancement based on training from examples. Particularly, in vision systems for aquatic robots, this training from examples allows the system to adapt the image restoration algorithm to the current environmental conditions and also to the task requirements. Image restoration involves the removal of some known degradation in an image. Traditionally, the most common sources of degradation are due to imperfections of the sensors, or in transmission. For the case of underwater images, additional factors are poor visibility (even in the cleanest water), ambient light, and frequency-dependent scattering and absorption, both between the camera and the environment, and also between the light source (the sun) and the local environment (i.e. this varies with both depth and local water conditions). The light undergoes scattering into the line of sight. the result is an image that appears bluish, blurry and out of focus.

Our approach is based on learning the statistical relationships between image pairs. In our case, these pairs are the image we actually observe and a corresponding color-corrected and deblurred images. This model uses multi-scale representations of the corrected (enhanced) and original images to construct a probabilistic enhancement algorithm that improves the observed video. This improvement is based on a combination of color matching correspondence with training data, and local context via belief propagation, all embodies in the Markov random field model. Training images are small patches of regions of interest that capture the maximum of the intensity variations from the image to be restored.

This work was done during my PhD while I was working in the Gregory Dudek. Examples of color restoration of underwater images

Experiment: Using different tones of poor colored images on the same training pair example. It shows that the algorithm is robust even with almost no color on the input image.

Please go to my list of publications for more detail on this subject.

Also, check out the Mobile Robotics Lab at McGill University for more info.