The following is all based on ideal elements and such; but it's where you start.
Gain of Inverting OPAMP: (R2/R3)
Gain of Non-Inverting OPAMP: [(R2/R3) + 1]
If we feed the "sensed" voltage into the inverting side, and a reference into the non-inverting side; then we have something like this: (Vref * ((R2/R3)+1)) - (Vsense*(R2/R3)).
To balance everything out, we make R3 = R1.
Next we need to know is how to pick R2 and R3. Based on the voltage divider chosen to sample the "sensed" voltage, the maximum voltage swing is roughly 0.66V. We want to maximize this swing at the output of the OPAMP. If we want a 5V swing from a 0.66V swing, that comes out to roughly a 7.5 voltage gain. Picking 10k for R3 and R1 (nice number, not too stiff, not too soft); means that R2 would have to be 75k to provide the 7.5 voltage gain on the inverting input (where the "sensed" voltage is connected to).
Now for the reference voltage. With the configuration as shown, we want to set the reference to match the maximum voltage at the "sensed" input. That way when we do the "Vref-Vense" it will start at 0V (or close to). In our example, this turns out to be around 2.49V. R_ref1 and R_ref2 were made the same to split the 5V reference and picked at 1300 to match R_bias.
Hope this helps. National Semiconductor, Texas Instruments, and Analog Devices have loads of information on this.
https://www.electro-tech-online.com/custompdfs/2010/11/AN-31-2.pdf
ADI - Analog Dialogue | Op Amp Applications Handbook