Dahua Confirms, Defends "Black", "White", "Yellow" Skin Color Analytics

Despite Dahua having repeatedly been caught with and then denying racial analytics, now Dahua not only confirms to IPVM it sells "black", "white", and "yellow" "skin color" analytics but resolutely defended them as being a "basic feature of a smart security solution".

However, virtually no other security or surveillance manufacturer offers such "skin color" analytics (though compare to Hikvision's Uyghur analytics).

Nonetheless, if Dahua, as one of the world's largest video surveillance manufacturers, makes this a "basic feature" as it says, this raises grave concerns about alarming and tracking people for being "black", "white" or "yellow", as determined by Dahua.

Dahua API Guide Offers "Skin Color", "Complexion"

Dahua's ICC Open Platform (ICC开放平台),, has an API guide for "human body characteristics" (人体特征) that includes "skin color" (肤色) / "complexion":

In a "data dictionary" (数据字典) Dahua explains that the three "skin color types" (肤色类型) it offers are "yellow" (黄), "black" (黑), "white" (白), plus "unrecognized" (未识别):

Skin color detection is mentioned in the "Personnel Control" (人员布控) category, a feature Dahua touts as part of its Smart Office Park solution intended to provide security for large corporate campuses in China.

Dahua Europe Cameras Offer Skin Color Detection

Dahua cameras listed in three different European countries also include "skin color" analytics in their specifications.

One is listed by Dahua Germany, another is listed by Dahua France, and three distributors in the Netherlands list a third (Bewakingswikel, PG Security Systems, and Osec.nl):

Note, "couleur de la peau" is French for "skin color". While each camera is currently listed, the specification sheets mentioning skin color analytics are dated 2021.

Dahua Defends Skin Color Analytics

In response to IPVM, Dahua strongly defended "skin color" analytics as "a basic feature of a smart security solution", arguing "black", "white", and "yellow" are only "three broad categories" that cannot be used to target racial groups:

The platform in question is entirely consistent with our commitments to not build solutions that target any single racial, ethnic or national group. The ability to generally identify observable characteristics such as height, weight, hair and eye color, and general categories of skin color is a basic feature of a smart security solution. As IPVM acknowledges, the API does not focus on skin color but includes it as one among 209 types of data enumerations, nor does it single out any specific skin color. The platform doesn’t even seek to create narrow skin color groups that could even theoretically be applied in a discriminatory fashion, but simply organizes three broad categories. As such, it is a simple mechanism to enable third-party developers to build smart solutions based on certain basic observable physical attributes. [emphasis added]

Note, Dahua's claim that it does not "single out any specific skin color" in its API is incorrect as it singles out three specific skin colors - "black" 黑, "white" 白, and "yellow" 黄.

Contrast To Previous Statements

Dahua has been caught repeatedly offering race-related analytics, as this video explains, but has never directly defended them before:

In past instances, Dahua denied and distanced itself from such technology, providing varying rationales:

• In November 2020, Dahua told The South China Morning Post it "does not sell products that feature [an] ethnicity-focused recognition function".
• In February 2021, following revelations of its "real-time Uyghur warnings", Dahua adjusted slightly, qualifying that "we never provided products or services for ethnicity detection in Xinjiang, PRC" [emphasis added].
• In an August 2021 post, Dahua changed its position again, alleging, "We have never developed a product or solution designed to identify or otherwise target any specific ethnic group."
• However, in an August 2022 letter to the FCC, Dahua admitted the authenticity of their "Uyghur warning" documentation, but alleged that it was a "historical internal software design requirement document" that was "never implemented"
• In October 2022, Dahua described "skin color" and "Xinjiang/Tibet" detection examples found by IPVM on its website as an "error", saying it was "deeply troubled", and emphasizing it had "never provided any customer of any type" with software that "targets" a racial or ethnic group.

Dahua's own current FAQ states: "We have never developed a product or solution designed to identify or otherwise target any specific ethnic group".

Color Analytics In Video Surveillance

While various video surveillance providers support color analytics for cars or clothing, virtually none offer this for skin color. Modern surveillance systems can, with significant limitations, attempt to analyze color.

Despite the technical possibility, analyzing skin color is overwhelmingly avoided. One of the most well-known cases was from a US startup Kairos (from outside the surveillance market) had offered what it called "diversity recognition" but shut it down 5 years ago, expressing concerns about biased results.

Technical Feasibility Problems

Even beyond bias concerns, skin color analytics suffer from various color fidelity problems as sunlight, shadows, and artificial lightning will impact the perceived or captured color by surveillance cameras.

More broadly, most cameras, including Dahua, default to using IR illumination in low-light conditions, which forces the camera into grayscale, creating even more difficult problems in identifying color.

Tracking Human Skin Colors Background

Companies can and do track human skin color in contexts outside security, e.g., for making diverse emoji icons or tracking skin cancer risk. The standard scale for doing this is called the Fitzpatrick Scale which uses six stages (named I through VI in Roman numerals):

Key differences include the Fitzpatrick Scale being based specifically on response to UV light, whereas Dahua uses visible light and has only three colors - "black", "white", and "yellow".

Dahua's version mirrors a concept in the PRC of there being "3 major races" (三大人种) as explained in an article in the government-run Chinese Academy of Sciences: black (Africans), white (Caucasians/Europeans), and yellow (Asians).

In the US and Europe, scientists began classifying people into separate 'races' based on skin color starting in the 17th century. However, this is now primarily considered as part of Scientific Racism.

Ethics Concerns

Race-tracking software elicits ethical concerns from experts over bias, discrimination, and profiling, as detailed in WSJ: The Quiet Growth of Race-Detection Software Sparks Concerns Over Bias.

That is why these analytics "are controversial and are rare outside of China" as IPVM explained in its Video Analytics Demographics Guide.

Within the PRC, skin color analytics are more common and even incorporated in PRC police standards (including one co-authored by Dahua.)

ACLU: Skin Color Analytics "Practically Guarantee" Racial Discrimination

ACLU Senior Policy Analyst, Jay Stanley, told IPVM that "very little good" can come from skin color analytics, and offering this "will practically guarantee" racial discrimination:

Very little good can come from training analytics to tag people by their skin color in video footage, especially here in the United States where racism is such a raw and deep part of our history and present. Video analytics already has the potential to lead to more pervasive surveillance; making detection of skin color a part of the toolbox will practically guarantee that racial discrimination becomes part of that. The history of "Uygur detection" analytics as part of the Chinese government’s infrastructure of oppression stands as a stark reminder of the ugly potential here. [emphasis added]

Security Industry Association

The Security Industry Association responded by raising concerns surrounding these analytics with two main concerns, including prejudice risks and privacy risks, IPVM:

While the firm in question is not a SIA member, our members are instructed to “oppose prejudice, harassment and abuse in the workplace as well as uses of their solutions that assist or abet prejudice, harassment or abuse.” The guidance is provided in the SIA Membership Code of Ethics which is affirmed when joining or renewing a SIA membership. The availability of such functionality gives greater opportunity for an individual or organization to use the solution to “assist or abet prejudice, harassment or abuse.” Furthermore, identifying an individual’s possible race (as skin color has often been used as a proxy for race) without their consent is guaranteed to generate data privacy and personal privacy concerns. On these matters of privacy, SIA further refers the industry to the principles outlined in the SIA Privacy Code of Conduct and the SIA Data Privacy Code of Practice for Video Surveillance. Both documents stress the importance of transparency and accountability, among other principles.

While Dahua is no longer a member of SIA, as SIA Expelled Dahua for ethics violations.