How close are we to solving the problem of automated visual surveillance?