Ab­stract

This paper pre­sents a method for the sim­u­la­tion of im­ages in the scope of vir­tual cam­era pro­to­types under the con­straint of color cor­rect­ness. This is a first step to gain a com­plete sim­u­lat­able cam­era model that can be used to gen­er­ate syn­thetic im­ages using real data. Each real cam­era sys­tem has its own color pro­cess­ing char­ac­ter­is­tics. Real im­ages recorded with a ref­er­ence cam­era model can be com­pu­ta­tion­ally sim­u­lated as if they have been recorded with an­other real or vir­tual cam­era. The re­sult­ing im­ages are trans­formed to un­derly the color char­ac­ter­is­tics of the tar­geted vir­tual cam­era sys­tem. Our ap­proach can be used at the de­sign phase of vi­sion-based ad­vance dri­ver as­sis­tance sys­tems to ver­ify the exact be­hav­iour under vary­ing op­ti­cal prop­er­ties of the op­ti­cal sys­tem and to test and eval­u­ate the over­all ro­bust­ness of the sys­tem when color pro­cess­ing changes. It can as well lead to a de­ci­sion basis for the se­lec­tion of the hard­ware to be used. In this paper, we show how cam­eras can be cal­i­brated and in a sec­ond step we eval­u­ate the sim­u­la­tion er­rors. Fi­nally, we apply our sim­u­la­tion to a traf­fic sign recog­ni­tion al­go­rithm and eval­u­ate its be­hav­iour in re­la­tion to ground truth data.