> I have some software which captures pics from my webcam at roughly 5 frames a second . 640 x 480 at 24bit
96 dpi hoz+vertical ( mouse right click properties win xp ) The software saves the *.bmp pics into a folder for me 901kb in size
i use the bload command to get the 24bit *.bmps into my freebasic prog ,
to load ok when fb is set to 16bit mode
ive not done a pixel colour analysis as yet between the original bmp file and what i see in the fb program window. Ill take a screen shot if its important and find out
FreeBASIC snoops the palette or whatever in the *.bmp file and does whatever it does or not ? to the pallete in fb prior to loading the data and displaying it as a picture
if i use bsave on the imported bmp and then bload the bsaved file the colours DO change. See prog below
Code: Select all
dim i as integer
'screenres 1300, 700, 15
screenres 1300, 700, 16
'screenres 1300, 700, 24
'screenres 1300, 700, 32
'screenres 1300, 700, ??
? "saved" :sleep
?"if in 15/16 bit mode colours change after loadeding back in using bsave 'saved.bmp',0"
? "saved again" :sleep
?"colours now seem to stay the same on after second load and save"
? " now doing test loop of saves and loads check for colour degredation "
for i = 1 to 60
I might have to bload , bsave and then bload each picture (sighs out loud @ 5fps+ ) in order to get the colours to stick. If you run the prog again but set fb to 24bit the colours seem to stick ( stay the same) all the way through
Im trying to capture pictures at the lowest useful size and color count and frame rate eg 640x480 16bit @ 5fps (possibly 12 bit colour would work) Im just using 1 camera
1024x768 32bit @ 10 fps (with multiple cameras some set at lower capture rates and sizes) would be better , but the computing costs are a tad high ......i am considering
a 2 cam system 1 lo res in B&W high frame rate @ 10-20 fps+ the other hi res at 640x480 @ 3-5fps
having captured at whatever specs, i then need to run fast analysis and also deep analysis and do so as close to real time as possible (which is why i have a computer network set aside for this task)
some of the analysis can be carried out on picture files that are LATER shrunk in size or reduced in color or both ( reducing the size of the files is important as ive got umpteen megabytes of data flying across the network )
Ive got around 5 freebasic gigaflops (see speedtest8.exe ) to throw at the vision tasks. I might be able to push this to 10 if there is no other way. Currently 3 P4s, 2 Dual cores & one Quad core dealing with just the vision.
The AI system runs for a couple of hours at a time and generates about a million files a session and terabytes of data which is sifted again post session before the decks are cleared.