Robotics Reference
In-Depth Information
Similar research has led to promising results elsewhere. At Intel, for
example, a group led by Ara Nefian has developed a system that could
identify four out of five words in noisy environments, and the results
were as good for Chinese as for English.
A Television Slave
One vision recognition technology that could find a wide audience is
Sharp's HiMpact Sports, which applies a set of algorithms that under-
stand what is happening in a game of Baseball, American Football or
Soccer, and can extract all of the exciting moments from a game, thereby
creating a summary of the video highlights without missing a single im-
portant moment in the game. The technology automatically indexes the
summary so that the viewer can see at a glance the whole list of high-
lights. It can also provide annotated summaries of the action. The re-
sulting summary permits sports fans to watch a game, play-by-play, with
a text commentary that will tell them what to watch for, which players
will be featured and the outcome of the play.
The software recognizes an event in a sport in all its permutations.
For example, it can distinguish different types of base hits in Baseball
(an infield grounder, or a line drive into centre field, or a bunt). The
summary allows the viewer to choose to watch all of the home runs scored
or all of a favourite player's hits during a game. This is achieved by using
a Hidden Markov Model (HMM), a technology that has previously been
successful in learning how to recognize spoken voices.
15
Just as an HMM
is “taught” words by training it on samples of different people speaking
the same word, Sharp's sports HMM was trained on video clips that it
categorized into different training sets, one set corresponding to a “play”
and one to a “not a play”. Thousands of video clips in these training
sets were flagged according to whether they constituted a play, and then
the HMM learned these clips, adjusting its internal parameters until it
could accurately classify those clips into plays and non-plays. Once it
had learned the two categories, the HMM was set so it could be used to
classify individual frames from a video as either part of a play or not.
Assisting Breast Cancer Diagnosis
One of the most important tools in the diagnosis of breast cancer is the
mammogram, an x-ray of the breast. A shadow on the x-ray might mean
15
See the section “Speech Recognition” later in this chapter.
