The problem may be inherent in the early state of face recognition at these rates. The Oak Ridge National Laboratory achieved more than 80 percent accuracy in a study that found faces through windshields, but it was at low speed. It may not be clear to anyone to crush a bridge.
Not that privacy publishers will necessarily think. Face Detection is already a questionable problem, let alone when it is used to look into cars. Whether you see it as an Orwellian burglary that can lead to abuse of power, it's at best accuracy issues. It sometimes has problems recognizing non-white people and women, and it is believed that a culprit will not wear a mask or other disguise. Although no terrorist detection system is foolproof, there are real concerns that current approaches can generate false positives or let suspects pass through the cracks.