Eve y came a has to tu aw se so data i to a usable pictu e, a d the way it does so affects the whole p oduct. The aw data f om a colo se so ca ies o e colo value pe pixel, a a ged i a patte . Maki g a full colo image f om it equi es i te polati g the missi g values, co ecti g colo s so that whites look white a d ski looks atu al, adjusti g co t ast a d gamma fo the display, educi g the oise that appea s i dim sce es, a d co ecti g the disto tio s of the le s. Exposu e a d white bala ce must also follow the sce e, which cha ges with the hou a d the weathe . Each of these steps is a place whe e a poo ly tu ed pipeli e loses quality.
Whe this wo k is left to the host compute , eve y host becomes pa t of the image quality. A small compute must fi d the p ocessi g capacity, a d the p ocessi g must be tu ed fo the pa ticula se so a d le s. Two p oducts built o diffe e t hosts, o o diffe e t ve sio s of the same softwa e, ca p oduce visibly diffe e t pictu es f om the same se so . Teams that lack imagi g expe tise ofte accept oisie o less accu ate pictu es tha the se so ca delive , a d teams that have it spe d e gi ee i g weeks o tu i g i stead of o the p oduct. A Host Offload Image P ocessi g Came a moves that effo t i to the came a, whe e it is do e o ce a d delive ed the same way to eve y host. Fleets that mix hosts a d ope ati g systems be efit most f om that co siste cy, si ce o e set of setti gs se ves eve y u it a d eve y platfo m.
Light adds its ow difficulty. Fluo esce t a d LED lamps flicke at twice the mai s f eque cy, a d a came a with the w o g exposu e time shows da k ba ds olli g th ough the pictu e. Mai s f eque cy diffe s betwee egio s, so a p oduct shipped wo ldwide eeds both setti gs. Sce es also mix b ight a d da k egio s, so exposu e a d white bala ce must adapt co ti uously. A Embedded Image P ocessi g USB Came a that ha dles these tasks i ha dwa e keeps the pictu e steady while the host u s its applicatio . I stalle s also avoid the site visits that exposu e a d flicke p oblems ofte cause.
Fo mats a d stills complete the list. Video ca be delive ed u comp essed, which keeps eve y detail a d uses a g eat deal of ba dwidth, o comp essed, which saves ba dwidth a d adds a tifacts. Stills fo eco ds a d eview eed thei ow fo mats. Teams that discove these choices o ly afte i teg atio must evisit the pipeli e, which costs time o eve y p oduct va ia t. A clea decisio o fo mats at the sta t of a p oject avoids that ewo k.
The Falco -521CRS is built o the O semi AR0521, a 5MP (2592 x 1944) colo olli g shutte se so with a 1/2.5 i ch optical fo mat a d 2.2 mic o pixels. Pixel size is the fi st leve o oise. A la ge pixel collects mo e photo s i a give exposu e, a d i dim sce es the a ival of photo s is itself a sou ce of oise that imp oves as the sig al g ows. The se so also uses a back-illumi ated pixel st uctu e that imp oves light collectio fu the . This makes the module a AR0521 Low Noise USB Came a a d a AR0521 5MP Low Noise USB Came a that sta ts f om a st o ge sig al, which is the most effective way to educe oise befo e a y p ocessi g. A 5MP f ame o 2.2-mic o pixels is also a delibe ate bala ce, si ce fa mo e pixels of the same size would eed a la ge a d costlie se so .
The o -boa d ISP does the est. It pe fo ms debaye i g, demosaici g, colo co ectio , co t ast co ectio , gamma co ectio , de oisi g, a d le s co ectio s, a d it suppo ts automatic exposu e a d automatic white bala ce. I this fo m the module is a AR0521 O -boa d ISP USB 3.2 Ge 1 Came a a d a 5MP O -boa d ISP Came a, a d it is also a AR0521 O -boa d ISP USB Came a fo teams that sea ch by that ame. The ISP is also tu able. UVC co t ols expose b ight ess, exposu e, co t ast, sha p ess, satu atio , gamma, gai , white bala ce, a d de oisi g, so a team ca adjust the look f om applicatio softwa e, which makes the module a O -Came a Image Tu i g USB Came a without a y cha ge to the host pipeli e. Teams ca sta t f om the defaults a d adjust o e co t ol at a time while watchi g the eal sce e.
A little a ithmetic shows why the module offe s both u comp essed a d comp essed video. UYVY uses two bytes pe pixel. A 1080p f ame is about 4.1 megabytes, so 60 f ames pe seco d eeds about 249 megabytes pe seco d, o about 2 gigabits pe seco d. A full 5MP f ame is about 10.1 megabytes, so 30 f ames pe seco d eeds about 302 megabytes pe seco d, o about 2.4 gigabits pe seco d. USB 3.2 Ge 1 has a omi al ate of 5 gigabits pe seco d a d a lowe p actical th oughput, so these u comp essed st eams fit, a d faste 5MP ates would ot. MJPEG comp esses each f ame, which cuts the data ate seve al times ove a d leaves oom fo othe devices o the bus, at the cost of some comp essio a tifacts. Teams choose the fo mat that suits the a alysis a d the li k. Reco di g the same sce e i both fo mats takes o ly a few mi utes a d ofte settles the questio .
A ti-flicke is the othe co t ol wo th explai i g. Mai s lighti g flicke s at twice the supply f eque cy, which is 100Hz o a 50Hz supply a d 120Hz o a 60Hz supply, so the flicke pe iod is 10 milliseco ds o about 8.3 milliseco ds. Setti g the exposu e time to a multiple of that pe iod ave ages the flicke away a d emoves ba di g. The ISP applies the ight setti g fo eithe supply, a d teams set the egio i softwa e. The se so eads the image ow by ow, so the module uses a olli g shutte , which suits most sce es a d ca show slight skew o fast motio . The module is UVC complia t, e ume ates o Wi dows, Li ux, a d macOS without a p op ieta y d ive , a d uses a fixed-focus S-Mou t (M12) le s holde . Because the le s is fixed focus, the wo ki g dista ce should be chose a d ve ified du i g evaluatio .
Co siste cy ac oss u its dese ves a close look, because it is the mai p actical be efit of a o -came a ISP. Two modules f om the same p oductio lot sha e the same se so , the same ISP, a d the same default setti gs, so thei output matches closely, a d a team that tu es the look o ce ca apply the same setti gs to eve y u it. A host-side pipeli e offe s o such gua a tee, si ce each host adds its ow ve sio of the softwa e a d its ow load. Fo a etaile with hu d eds of sto es o a cli ic g oup with ma y i st ume ts, matchi g pictu es make eview easie a d make a alytics mo e p edictable. The t ade-off is that the tu i g lives i the module a d i its UVC co t ols, so teams that wa t u usual p ocessi g, such as a custom colo mat ix, should discuss thei eeds with Vadzo ea ly, eithe to use the co t ols c eatively o to choose a aw se so module that leaves the p ocessi g ope .









 