Thanks for all the work jamesh! I'm sure we can get an open-source face-detection example written...jamesh wrote:It can also do face detection (not specific faces, but the sort of stuff you see on cameras), but I think that might need a licence as it's third party SW.
Not that it really matters, but, a lot of DSLR's don't do hardware stabilization in the lens but do it by physically moving the sensor (eg Sony and Pentax).jamesh wrote:Although the stabilisation is done in the GPU, it's done in software (detect shake movement, compensate for it). What I understand as hardware stabilisation would be more akin to that found in stabilising DSLR lenses, which requires support in the lens.
Yes, probably!Burngate wrote:So we could do image stabilization by attaching the camera module to a couple of motors driven off the Gertboard?
On that note - if there's anything I can do to help on the code side I'm glad to help. A Python driver/interface is probably right up my alley.jamesh wrote:You are not getting another word out of me....
Not necessarily true - see the Nokia 808 - that uses a 41MP sensor (same GPU as the Raspi) and produces the best pictures yet from a camera phone sensor. The preceding 12Mp on the N8 was also a very capable system.Stokestack wrote:I think the lower-rez camera is good news. 14MP on a puny chip is going to be a mass of useless noise, burying all that extra detail that people think they'll get. Noise also greatly reduces the effectiveness of codecs, as they waste already paltry bitrates trying to encode all of its spurious detail. Not to mention the extra work downscaling giant images for video.
A clean-as-possible 4MP is way more useful.
I think I know where you're coming from with the pixel count comparison, but even for low-light images the noise vs. Mpixel argument is really valid only with "all other things being equal" (eg. sense amp technology, and overall silicon die area) and as far as I know, it is not. The largest factor appears to be the total silicon sensor area, regardless of pixel count, and I suspect the latest high-megapixel sensors may also have the largest area. For a detailed technical discussion (including real data) see: http://theory.uchicago.edu/~ejm/pix/20d ... #pixelsizeStokestack wrote:A clean-as-possible 4MP is way more useful.
Code: Select all
Users browsing this forum: No registered users and 15 guests