|
In Unit 2, Professor Thrun says we can use the Kalman Filters to estimate another car's location and a noisy estimate of the velocity. However, that seems to be on the assumption that we know there are cars there. How do we detect objects? How do we know what it is we're detecting? And assuming we determine that there is an object and its a car, in practice how would you code the estimate for that particular object? |
|
Wild guess: On-board camera with machine learned object detection? |
|
Difference in laser scan around a certain area? You will know there is an object there, and with all kinds of machine learning software, you will eventually know it's a car? And velocity can be guesstimated, abrupt acceleration or braking is always possible of course... |
|
Usually you start with an image based feature detection algorithm like the the Scale Invariant Feature Transform algorithm (SIFT) or one of its brethern (SURF,ORB,etc). Then higher level algorithms for detection and tracking of moving objects (DTMO) are built from there. Here is an example paper that touches on some of these approaches and challenges in robotics. It may give you a better idea. |
