I do not agree with the PLL solution, because making that stable down to DC is very difficult if not impossible.
The best way to count low frequencies is the reciprocal methode.
You need 2 counters to do that.
One counter counts a high frequency reference during the gate time.
The second counter counts an integer number of the frequency to measure during the gate time.
As soon as the gate opens both counters wait for the leading edge of the unknown frequency. On this leading edge the real gate time (eg 0.1 seconds) will start. This time does not have to be very accurate, and can be done with a simple software loop. At the end of the gate time the counter will wait again for the leading edge of the unknown frequency. This will eliminate the traditional +-1 count error.
Now you know how many periodes of the unknown frequency and how many reference periods of the reference frequency were counted. Dividing the 2 and multiplying with the reference frequency will result in the exact frequency.
Accuracy is depends on reference accuracy and total gate time. The longer the gate time the more reference pulses are counted, the more accurate the result will be.