The performance of the face recognizer in VisualArea SDK depends on several factors, mostly:

• the quality of the source images
• lighting conditions
• number of people learnt
• number of identifications per person
• the angle of the face compared to the camera

When we did our testing we used 3 different videos having 3 diffent lighting conditions, all taken inside. One with normal dimmed natural backlight, one under neon light and one strong sidelights. Six people were trained within the system, their monograms are used as IDs. The videos show every face from various angles. 

Below you can see the test results on the frames of the video, but VisualArea SDK uses cummulated values, which eliminates the mistakes and strengthens the real results.

Here are some examples:

Recognition results of BB on the first video:

Recognition results of BB on the second video:

Recognition results of MM on the first video:

Recognition results of MM on the second video: