How face detection works

At the core of DISPL visitor analytics is advanced face detection technology, which relies on building a map of points on a face and finding specific areas or patterns. 

These points and their patterns are crucial for determining demographic parameters. The process is pretty straightforward: 

1. Sensor detects a visitor.
2. AI algorithm detects the borders of a visitor's face and builds a map of points on it.
3. AI compares areas and patterns of points to what it knows about facial features of different demographic groups.
4. The algorithm approximates the age and gender of the visitor.

This technology is fast, accurate, and reliable. However, these facial points must be clearly visible for it to perform optimally. In ideal conditions, with the face fully exposed and under good lighting, the system needs between 1.5 and 3 seconds to determine metrics accurately, depending on the version of the DISPL Kit used (newer is faster).

Impact of hijabs on analytics accuracy

Wearing a hijab can impact the time the system requires to identify these points. First of all, when a visitor wears a hijab, the detection time can increase. This is due to the partial coverage of the face, which limits the visibility of the facial points the algorithm relies on.

Various types of hijabs are worn differently across different countries, so it can affect the amount of the face covered and, consequently, the accuracy of the analytics.

We recognize that accounting for and testing every possible situation is challenging. However, DISPL CMS has a pre-built template displaying analytics metrics, which can easily be utilized to observe the results by showing different photographs to the sensor and analyzing the outcome.

Limitations in data accuracy

The limitation in the number of points that the algorithm can map on the face means the system might produce false results or incorrectly determine gender and age when the face is partially or almost fully covered. For partially covered faces, the accuracy is around 65%. While for almost entirely covered faces, accuracy drops even lower, sometimes not able to detect a face at all. When the face is fully covered or obscured by a mesh, such as in a burqa, the algorithm will not identify the visitor.

In addressing the compatibility of DISPL visitor analytics with visitors wearing hijabs, it's crucial to categorize hijabs based on the degree of facial coverage they provide, as this significantly affects the accuracy of facial recognition algorithms. 

Limited face exposure

The first group includes types like the Hijab and Shayla, which generally leave the face exposed, allowing for a limited, but operable accuracy rate of around 65% in gender and age detection.

Very limited face exposure

The second category encompasses the Niqab and certain styles of the Chador, which cover more of the face, only leaving the eyes visible, thus reducing recognition or even rendering detection impossible in some cases. Further tests on your side are needed.

No face exposure

The final group consists of the Burqa, offering the most extensive coverage, where the face is fully covered, including the eyes, with a mesh screen. In such cases, the DISPL's technology will not be able to identify the visitor due to the lack of visible facial points required for accurate analysis.

Additionally, specific metrics like hair color, hairstyle, and sometimes emotion cannot be determined for visitors wearing hijabs due to the lack of visible data points related to these attributes.

Conclusion

In conclusion, the DISPL visitor analytics solution relies on advanced face detection technology that maps facial points to determine demographic parameters accurately. While wearing a hijab may impact the accuracy of the analysis, DISPL's system can still provide insights for visitors wearing certain types of hijabs that expose most of the face. However, the accuracy drops significantly for more extensive facial coverings, such as those worn in the Niqab and Burqa. Still, DISPL CMS has pre-built templates to check the accuracy of the analysis, which can be utilized to test and check the results.

‍