Guest_Jim_* Posted July 28, 2015 Posted July 28, 2015 Eventually we may have highly capable robots in our homes, helping with chores and other tasks as we ask them, but before this can happen, they must be able to recognize the objects being used. To that end many objection recognition algorithms have been developed and work is ongoing to make ever better ones. Now researchers at MIT have combined object recognition with mapping algorithms to create a more capable system, without relying on special hardware. The researchers specialize in SLAM or simultaneous localization and mapping, which is a method for a robot to map their environment using multiple camera views. This generates a 3D map of the area and the objects in it using a single RGB camera. Normally an object recognition system using a single RGB camera works on single frames and relies on colors to distinguish objects from each other, so two objects near each other may be hard to tell apart, especially if they are similarly colored. As SLAM data includes depth information though, combining SLAM with object recognition makes it easier to distinguish objects from each other, and then identify. The researchers found their new method was able to compete with systems specially designed for objection recognition that use cameras capable of making depth measurements, like the Kinect, despite using a single, monocular camera. It also performs better outside, where the infrared light the Kinect uses is easily lost. Source: MIT Back to original news post Share this post Link to post Share on other sites More sharing options...
Recommended Posts