One of the most important of components of an AI which aims to replicate the Human intelligence is visual processing. There has been many improvements lately in Object recognition software. Detecting motion though, presents an extra layer of difficulty.

Scientists at Boston College have just developed a new method that lets a computer recognize a moving object with almost double the accuracy and ten times the speed of earlier methods.

The finding has just been presented at 2009 IEEE COnference on Computer Vision and Pattern recognition.

Previous methods relied on comparing the moving image with a databank of millions of images, which was a time consuming task with less accuracy. The new method relies on the technique used by the human eye, instead of searching through image databases with brute force. First, the rough location of the moving object is identified. Then many small regions are formed on the image of the object called thrust and search regions. These regions are constantly updated and compared with a memory bank by mathematical matches. To see original story visit the link here. www.physorg.com/news/164509831.html