Developmentoffeatureextractionengineforadvanced driver assistancesystems

Abstract

In this chapter, we demonstrated the experiments we performed to evaluate the pro- posedideaofdifferentiallyprivatecomputervisionalgorithms.First,toevaluatebLOM algorithm, we presented metrics to measure privacy and utility. Then we conducted differentexperimentstoshowtheeffectofbLOMalgorithmwhenitisappliedtodiffer- ent frequency bands. We showed that applying bLOM to mid-band achieved the best performance results. Then we showed the effect of changing block size inside the algo- rithm. We noticed that maximum performance of the algorithm was achieved whenwe settheblocksizetoavaluebetween5and10.Then,wemadeanend-to-endevaluation to demonstrate how facial recognition algorithm fails to detect person’s identity after applyingbLOMalgorithmwhilemaintainingtheabilitytodetectnecessaryfeaturesfor drowsiness detection algorithms like face, eyes and mouth. The final experiment was doneontheproposedyawningdetectionalgorithmwhichshowedthatcanachieve85% of privacy and utility after bLOM algorithm is applied to camera videoframes.