CMUCam control registers

[Don Reid]

I am writing (another) program to run the CMUCam and help set tracking parameters. (This one doesn't need Java, or Lab view). I have a couple of questions, that maybe someone can answer.

First has anyone (Kevin?) found a way to read out the sensor registers? I would like to see what the camera's internal code uses for exposure etc.

Second, I copied the various sensor settings from Kevin's code, and one in particular is causing trouble. Is anyone using anything other than the defaults.

Here is what I extracted from the code (maybe I got it wrong, any proof reading would be appreciated).

//Reg,  Value
{0x29, 0x80},	// Common I	= (AEC off)
{0x13, 0x20},	// Common B = (8 bit, auto func off)
{0x2D, 0x81},	// Common J = (flicker rates off)
{0x2A, 0x80},	// Frame Rate Adj 1 = (FRA on)
{0x2B, 0x20},	// Frame Rate Adj 2 = (-3.8%)
{0x2D, 0x81},	// Common J = (flicker rates off)
{0x12, 0x20},	// Common A (AGC on, AWB off)
{0x00, 0x00},	// AGC	= 1x
{0x01, 0x80},	// Blue Gain = 1 + 0
{0x02, 0x80},	// Red Gain = 1 + 0
{0x03, 0x80},	// Saturation = 1 + 0
{0x06, 0x01},	// Brightness
{0x10, 0x01},	// AE

(Also set NF to 0)

With the AE setting (last), I get an all purple image. With larger values (like 0x40), it works well. Since it is critical that this program use the exact same settings as the camera on the robot, I need to resolve this.

The plan is to have this program and our second camera to get near the field and check out the settings. We had a serious trouble with false positives at our first regional (Portland), and will be in Las Vega next weekend.

[Don Reid]

As a follow up, at the Las Vegas regional, I was able to test several different AE settings, and found that at 1, it was very hard to detect the target. Just a little higher, 4, brings the target to a much higher level, and still doesn't require accepting saturated values (240). With an AE setting of 4, we see the target well with:

Red (Cr) 35 - 59
Green (Y) 149 - 205
Blue (Cb) 123 - 149