Testing of detection systems

The phrase “probability of detection” (P_d) is often misused in the context of exterior electronic perimeter intrusion detection systems (PIDS). For example, a system which correctly detects when tested three times has detected at a rate of 100%, but might not have a “probability of detection” of 100%. “P_d” is often misused to describe this experimental result as a 100% probability of detection. “Confidence limits” for P_d can be estimated and minimum extents for testing can then be demanded. These confidence limits are sometimes calculated using the normal, or the chi-squared, statistical distributions. Using these distributions for PIDS testing is questionable because they are often used outside their region of validity. There are fundamental constraints which obstruct conventional statistical confidence limits for P_d from being found. P_d statistics are discussed and an exact way to express detection uncertainty using the binomial distribution is described. Ways to reduce testing, yet maintain confidence, are discussed. Experts are concerned that extensive, but unrepresentative, testing gives an unwarranted impression of reliability. This paper emphasises the importance of representative testing and the clear presentation of results