Decoding human behaviour is a tough business. Anyone who works with human data to make decisions knows that the quality of the output starts with the quality of the input. In click data, the challenge is to remove bots and false clicks. In claimed data, the challenge is to remove poor participants. Biometrics is no different except for one distinct advantage. The machine can observe real live human beings, in real-time, and quickly identify the bad actors.
This not only cleans up biometrics data. It also helps clean up claimed and click data.
Here are some of the quality controls we use to make sure that your data inputs produce the highest quality outputs:
1/ Face Detected - first, we always look to see if a face is detected before understanding if expressions are present.
2/ Colour Value - many face detection algorithms do not work on or exclude darker skin tones. Our model specifically recognizes and excludes pixels that are not part of the face providing much higher quality outputs for darker skin tones while providing information about the environment.
3/ Jittering - we are able to determine if someone is moving too quickly for a read and thus remove false measures
4/ Facial Feature Proportions - we simultaneously detect how much of the eyes and face are inside the frame to understand the quality and confidence behind the data
5/ Video Quality - we detect a number of measures around frames per second and bitrate to assess the quality of the video and resulting measures of attention, expression, and emotion.
6/ Successful Calibration - this is a very easy way to determine if someone is paying attention by simply completing a dot clicking exercise.
Our systems use the above criteria to set thresholds and determine the quality level of your data. In some cases, we do not allow participant data to be included. In others, it is up to you.
We also share these metrics with our partners to quickly identify poor participants and help determine better bid prices.
As we continue to expand our capabilities, we continue to unlock ways to better the quality of the data resulting in better quality experiences from that data.