Existing optical scan voting systems depend on the integrity of the scanner. If a compromised - or merely faulty - scanner reports incorrect results, there is no ready mechanism for detecting errors. While methods exist for ameliorating these risks, none of them are entirely satisfactory. We propose an alternative: a radically open system in which any observer can simultaneously and independently count the ballots for himself. Our approach, called OpenScan, combines digital video recordings of ballot sheet feeding with computer vision techniques to allow any observer with a video camera to obtain a series of ballot images that he can then process with ordinary optical scan counting software. Preliminary experimental results indicate that OpenScan produces accurate results at a manageable cost of around $1000 in hardware plus $0.0010 per ballot counted.