How AI reads the writing on the road
In this video, Nvidia is going to show you how this DNN has evolved to detect not just Lane lines, but also other important road markings and visual landmarks.
It is now called Mac net, and in this clip, we will see results from a map net DNN model that performs complete end-to-end detection of road boundaries shown in white dash line, lines in blue, solid lane
lines in red, as well as vertical landmarks such as poles in yellow map net also detects intersection entry and exit lines in orange road markings, such as HOV Lane signs in cyan and cross traffic intersection lines in magenta.
Here map net is detecting road markings and landmarks on a Japanese highway including arrows road text on the left lane denoting an upcoming exit and road text on the right lane that indicates to keep straight to stay on the highway.
And here map net is detecting road text on the right lane that denotes an upcoming merge as well as detecting vertical guideposts in yellow, where these guideposts end we also observe a highly accurate transition from solid to dashed lane line detection.
Map net now also enables camera based high quality map creation and localisation using commercially available sensors for consumer vehicles.
For more information, watch the video below.