Functional safety aspects of pattern detection algorithms

Pattern detection algorithms may be used as part of safety-relevant processes employed by industrial systems. Current approaches to functional safety mainly focus on random faults in hardware and the avoidance of systematic faults in both software and hardware. In this paper we build on the concepts of the international standard for functional safety IEC 61508 to extend safety-relevant notions to numerical and logical processes (algorithms) employed in pattern detection systems. In particular, we target the uncertainty pertaining to face detection systems where incorrect detection affects the overall system performance. We discuss a dual channel system that comprises two of the most commonly used and widely available face detection algorithms, Viola-Jones and Kienzle et al. We present a method for deriving the probability of failure in demand (PFD) from the combination of these two channels using both: 1oo2 and 2oo2 voting schemes. Finally, we compare experimental results from both the perspectives of availability and safety, and present conclusions with respect to the appropriate choice of information combination schemes and system architectures.