circle detection and read time
in [Matlab]
|
Prev: Video mosaicing
Next: break/return upon user input
From: Julia Woody on 9 Apr 2010 08:32 a) please describe a strategy for an algorithm that locates a circular watch and reads time from analog clocks/watches automatically. b)Suppose the watches are not positioned upright, they are arbitrarily rotated. Is it possible to read time in this case? If yes, how? (hint: consider the watch bands) thanks Julia
From: Dave Robinson on 9 Apr 2010 10:05 "Julia Woody" <jwwoody0(a)gmail.com> wrote in message <hpn6o5$s1m$1(a)fred.mathworks.com>... > a) please describe a strategy for an algorithm that locates a circular watch and reads time from analog clocks/watches automatically. > > b)Suppose the watches are not positioned upright, they are arbitrarily rotated. Is it possible to read time in this case? If yes, how? (hint: consider the watch bands) > > thanks > Julia 1) Use a circular Hough transform (available in the FEX ) to identify the watch face 2) Use a line Hough transform (available in IP toolbox) to identify hands (which lie inside the watch face circle. 3) Line Hough gives length of line, so you can differentiate hour from minute hand, also gives line angle which gives you the time. Regards Dave Robinson
From: Walter Roberson on 9 Apr 2010 11:22 Julia Woody wrote: > a) please describe a strategy for an algorithm that locates a circular > watch and reads time from analog clocks/watches automatically. > b)Suppose the watches are not positioned upright, they are arbitrarily > rotated. Is it possible to read time in this case? If yes, how? (hint: > consider the watch bands) The answer for (b) is NO. Watch bands is a red herring. *Some* watches have watch bands, but not all analog watches do. For example, "pocket watches" do not. It is common for watches to have stems, and around here it is traditional that the stems for wrist-watches to be at or near the "3 o'clock" position, but I have seen analog watches that had no stem. (You may perhaps be wondering how the time was set on those watches, but on that matter, you have to keep in mind that "analog" as applied to watches only refers to the use of hands to display the time, and not to how time is internally maintained by the watch. Adjusting the time can be handled by buttons anywhere on the watch, or by dials anywhere on the watch, including on the back of the watch or inside the watch cover.) You may perhaps have thought of the possibility of doing OCR (Optical Character Recognition) to determine the position and orientation of the numbers on the watch. The problem with that is that not all analog watches have numbers. Braille watches for example. And I have seen a "fashion" wrist-watches that only had four "tick marks", corresponding to the 12, 3, 6, and 9 positions, with the user expected to use their knowledge of angles to interpolate the intermediate hours. I have also seen pictures of expensive watches that had no numbers or hands, but which had dots which light up, the outer circle of dots for the hour, the inner circle of dots for the minutes. Was that an analog watch or a digital watch? It had no _digits_ on it, so it would be hard to claim it was a digital watch...
From: Alan B on 9 Apr 2010 12:01 Walter Roberson <roberson(a)hushmail.com> wrote in message <hpngnb$gvd$1(a)canopus.cc.umanitoba.ca>... > Julia Woody wrote: > > a) please describe a strategy for an algorithm that locates a circular > > watch and reads time from analog clocks/watches automatically. > > b)Suppose the watches are not positioned upright, they are arbitrarily > > rotated. Is it possible to read time in this case? If yes, how? (hint: > > consider the watch bands) > > The answer for (b) is NO. Watch bands is a red herring. *Some* watches > have watch bands, but not all analog watches do. For example, "pocket > watches" do not. It is common for watches to have stems, and around here > it is traditional that the stems for wrist-watches to be at or near the > "3 o'clock" position, but I have seen analog watches that had no stem. > (You may perhaps be wondering how the time was set on those watches, but > on that matter, you have to keep in mind that "analog" as applied to > watches only refers to the use of hands to display the time, and not to > how time is internally maintained by the watch. Adjusting the time can > be handled by buttons anywhere on the watch, or by dials anywhere on the > watch, including on the back of the watch or inside the watch cover.) > > You may perhaps have thought of the possibility of doing OCR (Optical > Character Recognition) to determine the position and orientation of the > numbers on the watch. The problem with that is that not all analog > watches have numbers. Braille watches for example. And I have seen a > "fashion" wrist-watches that only had four "tick marks", corresponding > to the 12, 3, 6, and 9 positions, with the user expected to use their > knowledge of angles to interpolate the intermediate hours. > > I have also seen pictures of expensive watches that had no numbers or > hands, but which had dots which light up, the outer circle of dots for > the hour, the inner circle of dots for the minutes. Was that an analog > watch or a digital watch? It had no _digits_ on it, so it would be hard > to claim it was a digital watch... Maybe that would be a bital watch?
From: albert herman on 26 Apr 2010 09:28
"Dave Robinson" <dave.robinson(a)somewhere.biz> wrote in message <hpnc75$p8a$1(a)fred.mathworks.com>... > "Julia Woody" <jwwoody0(a)gmail.com> wrote in message <hpn6o5$s1m$1(a)fred.mathworks.com>... > > a) please describe a strategy for an algorithm that locates a circular watch and reads time from analog clocks/watches automatically. > > > > b)Suppose the watches are not positioned upright, they are arbitrarily rotated. Is it possible to read time in this case? If yes, how? (hint: consider the watch bands) > > > > thanks > > Julia > > 1) Use a circular Hough transform (available in the FEX ) to identify the watch face > 2) Use a line Hough transform (available in IP toolbox) to identify hands (which lie inside the watch face circle. > 3) Line Hough gives length of line, so you can differentiate hour from minute hand, also gives line angle which gives you the time. > > Regards > > Dave Robinson Hi all, Could you please give your ideas. My question is that : Suppose the circular watch image is rotated by 90 degree clockwise and the diameter of the circular part is K1. How to find an affine transformation parameters that transforms the image to an upright clock image with the diameter of the circular part as K2 Also, what do you think that what kind of modifications do you need to make to detect and read rectangular shaped watches with arbitrary orientation? |