I take substantial amount of my pictures in low light conditions. However, the autofocus of my K20D has proven to be a poor companion under these circumstances. I usually have to fight with problems of getting the focus right. The problem is made even worse by the fact that the camera triggers the AF light of the flash far too late. There is some brightness in which the camera can’t focus quickly or not at all but doesn’t attempt to use the flash’s AF illuminator. I have remedied this condition with the development of several AF illuminator assistants, basically a second- manually triggered- AF-illuminator. Although I used pretty strong light sources in them, performance was still sub par. After reading about the “Pentax yellow light focus problem” that Ricehigh described, I decided to do some tests myself.
Basically what is known from other Pentaxians is that under incandescent illumination a front focus of the camera can be observed. It has been speculated that this is related to the amount of infrared radiation contained in this light. Therefore I tested whether the Pentax AF reacts to near IR or UV light. The good news: no it does not at all. I tried UV LEDs and very near IR LEDs, just at the border of human perception, and with both I could not get any AF lock-on. I then testes a range of LEDs with comparable brigthness on a focus test chart (please note my excellent hand drawing skills ). Here are the results (left is front, right is back, 45 deg angle, focus until stable, two pictures each, both where always exactly the same (every time refocused)):
What you can see is that there is very little dependency on wavelength. For none of the used wavelengths, IR, UV, red, green and blue (not all are shown) were there siginificant differences in accuracy, if the focus could get a lock-on (not for UV and IR) at all. Therefore each color should be able to do the trick in an autofocus assist light.
Then I did a second test: I again mounted the camera on a tripod and aimed it at a lamp in my dark living room. Now I illuminated the lamp with various LEDs and tested how well the focus worked. Here are some examples:
As you can see, each color works. Surprisingly, even the dim blue LED can get a lock-on (in terms of energy it is as bright as the red and the blue). The surprising fact however was that the bright red LED, which even projects an image of its die onto any surface (really short focal length of the LED dome lens) caused the most problems. The focus is able to lock on but often it was not accurate. Just as it is really difficult to manually focus under red light- due to the low acuity of the eye for red- the camera had the same problem. Often the focus was slightly of to the front or back. The best results were obtained wth the green LED. Now I understand why the Fuji bridge camera of my friend Jonas uses a green LED (besides the fact that in LiveView green is alwazs superior due to twice as manz green pixels)! I will try to test a green LED for my autofocus assist light as well! So please come back because I will present this neat little device shortly on my site!
What do we learn from all this? First, the K20D does not appear to necessarily show a problem with the autofocus for different wavelengths. At least not if the illumination is decent. This excludes a systematic error in the AF and makes a “yellow problem” unlikely. I have also never experienced one. For near threshold intensities it seems to be that green light can be best used for accurate AF. Lock on is fastest and most of the times more accurate than red light. Another advantage is, that green LEDs are still effective and bright and green light does not trigger the pupil reflex as badly as blue light.
This test is not really “scientific” as there are no accepted standards for this kind of measurements. Since the spectral response curve of the AF sensor and filters in front of it are unknown, no equal luminance setup can be built. It is merely a test which just tests if there are any grave problems with the AF under different wavelengths and which of those could best be used for an AF assist lamp.