Predator: Tracking + Learning

Zdenek Kalal has come up with a system that can quickly learn how to track objects in a video stream. The demo is quite interesting. The related papers are mentioned at the end of the video (and also on Zdenek's website).

To decide if the tracking algorithm is actually so robust or not, I'll have to go through the papers. But for now, the video is quite impressive. Oh, by the way, this isn't an april fool's video ;)

OpenCV Face Detection Visualized

This video visualizes the detection process of OpenCV's face detector. The algorithm uses the Viola Jones method of calculating the integral image and then performing some calculations on all the areas defined by the black and white rectangles to analyze the differences between the dark and light regions of a face. The sub-window (in red) is scanned across the image at various scales to detect if there is a potential face within the window. If not, it continues scanning. If it passes all stages in the cascade file, it is marked with a red rectangle. But this does not yet confirm a face. In the post-processing stage all the potential faces are checked for overlaps. Typically, 2 or 3 overlapping rectangles are required to confirm a face. Loner rectangles are rejected as false-positives.

New statistical model of vision

The human retina is made up of approximately 100 million light sensitive cells. This is enough to overload all neurons in the brain. So, people believed that the brain somehow reduced this information overload - by interpreting things in terms of horizontal, vertical and diagonal lines. Then, combining these lines and edges into objects that could be recognized.

But recently, Ruth Rosenholtz put forward a model that this happens only near the center of the retina. At the periphery, things a quite different. The brain does not interpret signals from the periphery as good as the center - a phenomenon you can feel physically (there's a test at the bottom link).

Earlier, this was just a side comment. There wasn't any mathematical model to account for this peculiarity. This new model of vision is converting that into numbers and calculations!

[Source]

Tic Tac Toe + OpenCV

Here's an interesting video I watched on YouTube today. It has sparked some interesting ideas, similar to the SuDoKu Grabber. Lets see if I can create something!

[Source]

OMG! She kissed him!

The people at the University of Oxford have developed a computer vision based system that detects interactions between people. The system can recognize four gestures as of now: Hand shakes, hugs, kisses and high-fives.

It uses an upper body detector and then uses cues like the orientation of the head, relative distance between people, etc

The Active Vision Group and the Visual Geometry group

Matlab now supports GPGPU

The latest version of Matlab now supports General Purpose GPU. The Parallel Computing toolbox comes with support for nVidia CUDA devices. [updates]

Self-recharging robots

Several companies are working on taking recharging to the next level - Robots that locate wall sockets and plug themselves in. The robot has several sensors and an arm that locate the socket and plug into the socket.

This got me wondering, can there be a machine-vision-only solution to this problem? A socket should be a very easily distinguishable thing. Hmm... could the Scale Invariant Feature Transform do this trick? Or will simple template matching work?

[Source]