Following last week's story about the police and the DVLA's number plate-checking experiment, it turns out that the driving authority's data is only around 40 per cent accurate. The information came to light in a report into the year-long trial, which occupied 23 police forces using new automatic number plate recognition (ANPR) technology.
The plan is to detect crims by cross-checking the identity of the driver against the DVLA and police databases.
The report, commissioned by Home Office and carried out by PA Consulting, found that the quality of the data worsened as the trial progressed. Apparently, the police had to resort checking on tax discs visually rather than relying on DVLA data because of "decreased confidence in using the DVLA databases as the primary means of stop".
The report went on, "The DVLA databases in particular were shown to be poor. As yet, there are no Criminal Justice Extranet facilities for the electronic transfer of updated information to forces on a daily basis and this is a further limitation on database accuracy. Further, there is a lack of rigorous understanding as to the precise causes of data inaccuracies. As data is key to ANPR, we conclude that this represents a weakness that should be addressed."
Recommendations made by the report into ways of improving accuracy include more linking of police and MOT databases and a "vehicle intelligence data warehouse " -- although this sounds a lot like a duplication of what the DVLA is supposed to do now.
"This data warehouse should also hold ANPR reads and hits, which are themselves a vital source of vehicle intelligence and should be accompanied by the development of data mining tools of a more sophisticated nature," the report went on.
However, a DVLA spokeswoman claimed that, since the report appeared, accuracy has improved massively to some 92 per cent.
For those keen on retaining some privacy, the report points out the technical difficulties of making databases talk to each other, and that funding of the project remains an issue.