eqqmc2
03-19-2005, 06:09 AM
I posted this message at dpreview and at networks54.Figure it could help here too:
After seen so many threads on this subject on a variety of prosumer digital cameras forums I did some research on the patented autofocus mechanisms that Minolta has put out. As many of you know for passive autofocusing (ie no AF assist IF light) the two most common mechanism/algorithms are the phase detect method patented by Honeywell, in 1980 (Minolta was sued and had to pay a lot of money for illegally using this patent) and contrast detection method. The phase detection method uses at least two AF detectors and assumes that a focused image is when the light that passes through the two separate paths are in phase. This is a more expensive and involved mechanism and Minolta use it primarily on their SLR cameras (anyone knows of any technical reason??). There are a few patents from Minolta detailling their designs. Contrast Detection method for the continous AF mode is documented in patent No 6636263
http://www.freepatentsonline.com/image-6636263-1.html or
http://www.freepatentsonline.com/patents/us/663/6636263/6636263.pdf
This patent and others are from the middle 1980's to 2001.
The algorithm is pretty simple: for the AF area (which is a subset of the total CCD area, if anyone knows whats is this AF area is (No of pixels) for the A200 or A2 that would be very helpful)calculate:
Every pixel (i,j) has intensity X(i,j) where i=column number and j=row number of the AF area
For every pixel location i,j in AF area do
Build the matrix:
Delta(i,j)= X(i+1,j)-X(i,j)
enddo
Sum over all pixels in AF area:
Xn = Sum((Delta(i,j),i=1,N,j=1,M)
Xn = what Minolta refers to as the AF Evaluation Value which is nothing more than a Focus Measure
Now, the AF mechanism moves the lens at small steps increments (computing Xn at each step) and determines the position that has the MAXIMUM Xn.Thus, n is the position of the lens at every step. This is the position where it is assumed that the image is focused. The lens movement happens when you depress the shutter halfway. The location of maximum Xn happens when the locked AF signal shows on the EV or LCD.
If you carefully notice, the matrix Delta(i,j) is nothing more than the vertical contrast at each pixel location with respect to the right neighbour pixel. The patent doesnt specify which component (R,G,B) of the light intensity is used. Someone has already offered that it could be the green component but I am not sure. Any comments here ???
As you can see, this type of AF WILL NOT FOCUS IN LOW LIGHT unless you are focusing in an area that has some small light source with enough high relative intensity and will focus correctly in LIGHTED SCENES only if you have vertical contrast in you AF area, ie vertical lines of relative bright and dark fringes. I have tested this algortithm and it is true all the time.
Solution for LOW LIGHTING: I have tried a laser pointer which has different patterns (red light vertical contrast) and gotten excellent results in focusing the A200 in TOTAL DARKNESS (using my 5600HD flash or built in flash). I will post the details if I see a few requests).
As for the comments on IF AF Assist: To make this systems work you must have an IR sensor that measures the time that it takes for the IR beam to travel to the subject and back to the camera. Knowing the time and speed of the IR light the object distance is computed. This mechanism is also documented in several Minolta patents. The A200 does already have an IR sensor, however, this doesnt mean thats is located in the proper area for the computations of the time. So even if minolta would put out a firmware upgrade to use the AF assist light of the 3600 or 5600 there could be some other problems such as deciding how to optimize the IR focused image with the vertical contrast algorithm already used in the camera and the additional logic required to make it work with the flash. Here is note to those "bright" Minolta business decision makers: if you had built the a200 or a2 so that it could have used the IR assist mechanism in the 3600 or 5600 (but only thru these flashes), you would have force us to purchase those flash units, creating a great product synergy and increasing the revenue needed at such a bad economic position as KM is at right now.
Last note, if there are any minolta engineers reading this forum, as they should be, please bring out some coments and some more details. I am not too impressed with your engineering judgement (of the AF system only, everything else is AWSOME) of this camera design and if i am mistaken please point it out. Also why not implement horizontal contrast design?? Why remove the wireless flash capability that existed in the A2.
It is a shame, cause this camera and the A2 could have been the best prosumer cameras in the market.
Hope this helps!!
After seen so many threads on this subject on a variety of prosumer digital cameras forums I did some research on the patented autofocus mechanisms that Minolta has put out. As many of you know for passive autofocusing (ie no AF assist IF light) the two most common mechanism/algorithms are the phase detect method patented by Honeywell, in 1980 (Minolta was sued and had to pay a lot of money for illegally using this patent) and contrast detection method. The phase detection method uses at least two AF detectors and assumes that a focused image is when the light that passes through the two separate paths are in phase. This is a more expensive and involved mechanism and Minolta use it primarily on their SLR cameras (anyone knows of any technical reason??). There are a few patents from Minolta detailling their designs. Contrast Detection method for the continous AF mode is documented in patent No 6636263
http://www.freepatentsonline.com/image-6636263-1.html or
http://www.freepatentsonline.com/patents/us/663/6636263/6636263.pdf
This patent and others are from the middle 1980's to 2001.
The algorithm is pretty simple: for the AF area (which is a subset of the total CCD area, if anyone knows whats is this AF area is (No of pixels) for the A200 or A2 that would be very helpful)calculate:
Every pixel (i,j) has intensity X(i,j) where i=column number and j=row number of the AF area
For every pixel location i,j in AF area do
Build the matrix:
Delta(i,j)= X(i+1,j)-X(i,j)
enddo
Sum over all pixels in AF area:
Xn = Sum((Delta(i,j),i=1,N,j=1,M)
Xn = what Minolta refers to as the AF Evaluation Value which is nothing more than a Focus Measure
Now, the AF mechanism moves the lens at small steps increments (computing Xn at each step) and determines the position that has the MAXIMUM Xn.Thus, n is the position of the lens at every step. This is the position where it is assumed that the image is focused. The lens movement happens when you depress the shutter halfway. The location of maximum Xn happens when the locked AF signal shows on the EV or LCD.
If you carefully notice, the matrix Delta(i,j) is nothing more than the vertical contrast at each pixel location with respect to the right neighbour pixel. The patent doesnt specify which component (R,G,B) of the light intensity is used. Someone has already offered that it could be the green component but I am not sure. Any comments here ???
As you can see, this type of AF WILL NOT FOCUS IN LOW LIGHT unless you are focusing in an area that has some small light source with enough high relative intensity and will focus correctly in LIGHTED SCENES only if you have vertical contrast in you AF area, ie vertical lines of relative bright and dark fringes. I have tested this algortithm and it is true all the time.
Solution for LOW LIGHTING: I have tried a laser pointer which has different patterns (red light vertical contrast) and gotten excellent results in focusing the A200 in TOTAL DARKNESS (using my 5600HD flash or built in flash). I will post the details if I see a few requests).
As for the comments on IF AF Assist: To make this systems work you must have an IR sensor that measures the time that it takes for the IR beam to travel to the subject and back to the camera. Knowing the time and speed of the IR light the object distance is computed. This mechanism is also documented in several Minolta patents. The A200 does already have an IR sensor, however, this doesnt mean thats is located in the proper area for the computations of the time. So even if minolta would put out a firmware upgrade to use the AF assist light of the 3600 or 5600 there could be some other problems such as deciding how to optimize the IR focused image with the vertical contrast algorithm already used in the camera and the additional logic required to make it work with the flash. Here is note to those "bright" Minolta business decision makers: if you had built the a200 or a2 so that it could have used the IR assist mechanism in the 3600 or 5600 (but only thru these flashes), you would have force us to purchase those flash units, creating a great product synergy and increasing the revenue needed at such a bad economic position as KM is at right now.
Last note, if there are any minolta engineers reading this forum, as they should be, please bring out some coments and some more details. I am not too impressed with your engineering judgement (of the AF system only, everything else is AWSOME) of this camera design and if i am mistaken please point it out. Also why not implement horizontal contrast design?? Why remove the wireless flash capability that existed in the A2.
It is a shame, cause this camera and the A2 could have been the best prosumer cameras in the market.
Hope this helps!!