Depending on quality of enrolled images and the face images you have coming off your cameras, as well as other factors, your results with safr may generate too many false matches.  You can address this by improving enrolled image quality (avoid low resolution and especially occluded faces) as well as some other tips laid out below.

First, consider if Are you differentiating between the light beluw "partial matches" vs. the light green "100% matches"?
Here is a partial match.  
See thi table on this page:  For info about how to interpret the % values.  As you can see, 94% should only be considered a possible match.  If you only want to see 100% positive matches, directions are shown how to do this below.

Disable Mask Recognition

Secondly, if you are finding many false matches with occlusion, try to disable mask recognition as follows:

  1. Go to Video Feeds window

  2. Chose "Advanced Feed Configuration".  You will see something like the following:

  3. Click "Add" in upper right

  4. Search for "mask" and click "recognizer.detect.mask" and click Add button

Do this for each feed.

If you wish to not get partial match reports, set the proximity threshold to 0.  Use the same process as above but set property "recognizer.idenity-proximity-threshold-allowance is as follows:

Diable recognition for low quality enrollment phots

Finally, watch for the faces that results in false matches often.  Work on getting an improved face images for those persons. In the meantime, create a "No match" group and put them in for time being to stop the false matching.

Like this:

Creat "No match" group:
1. In people window, select "..." in upper right and click "Manage Groups"
2. Click "+" to add a group.  Then give group name "Needs better image" and uncheck "Enable face recognition"

When you find a bad match in events window, chose "view person activity" as follows:

Click the edit pencil button:

In the person record, find the groups field and click the Edit button (pencil icon) and select the "Need better image" group.