MIT researchers develop new algorithm for improving object recognition




Researchers at MIT have developed a new algorithm that object recognition could improve and accelerate. The code is particularly useful for robots that are used for example to perform simple household tasks.

When the identification of objects, the probability of a correct recognition by an algorithm considerably higher when an image sensor shooting images from multiple angles, instead of only one viewing angle. From this data MIT researchers have adapted an existing algorithm that, among other things is used in radar systems. In addition, the algorithm generates more ideas on what can be an object by comparing this hypothesis with other images, writes MIT.

To the recognition mechanism not to be too slow, only held the most probable hypothesis, while the less probable identifications are deleted. Nevertheless, the algorithm is still quite slow because the recently adopted hypothesis still needs to be checked.

The MIT researchers came up with the idea of ​​the algorithm would be to amend that not all assumptions are removed, but only part. For the selection, samples are randomly selected from the total. Because there are on average much overlap in the hypotheses arises eventually consensus on what the object is very likely, with a certain number of samples. Through some clever methods of comparison with images from different angles, the algorithm would be significantly faster than the “old” method.

According to the researchers, the algorithm can be used for good robots must be able to recognize objects quickly. It is especially during trials looked at objects that are found in a household.


In: Technology & Gadgets Asked By: [15464 Red Star Level]

Answer this Question

You must be Logged In to post an Answer.

Not a member yet? Sign Up Now »