德里警方招标 338 台人脸识别摄像头，覆盖共和国日与独立日路线
Delhi Police Sought Facial Recognition CCTV at Ramlila Maidan
2026 年 7 月，德里警方就中央区和新德里区 338 台 CCTV 摄像头招标，招标文件要求固定摄像头和 PTZ 摄像头具备人脸识别能力，并配备人群聚集、遗留物检测和车牌识别分析功能，招标估值 15.66 亿卢比，8 月 20 日截标。文件称摄像头用于覆盖共和国日和独立日庆祝活动的 VIP/VVIP 路线及其他地点，并要求视频管理软件支持人脸识别系统、厂商提供可查看最多 500 路实时视频的移动应用。招标文件未设定人脸识别准确率标准、比对数据库、相似度阈值、检索权限或人脸数据留存期限，全文仅一次提及隐私，且未引用《数字个人数据保护法》或 Puttaswamy 判决。
正文
当前显示原文，中文全文译文尚未就绪。
.entry-header
Access the copies of the documents: Request for Proposal | GeM bid document | Technical specifications
In July 2026, Delhi Police invited bids for 338 CCTV cameras across its Central and New Delhi districts. The specifications require facial recognition on fixed and pan-tilt-zoom (PTZ) cameras, alongside analytics to detect crowds, abandoned objects and vehicle number plates, according to the tender documents reviewed by MediaNama. The Internet Freedom Foundation (IFF) first reported on these tender documents.
The request for proposal (RFP), issued by the office of the Additional Commissioner of Police (Provisioning & Logistics), says the cameras are meant “to cover VIP/VVIP route & other locations” for the Republic Day and Independence Day celebrations. Delhi Police estimated the tender’s value at Rs 15.66 crore. The bidding process closed on August 20.
The documents reveal that Delhi Police intends to equip every type of camera with facial recognition capabilities:
Fixed cameras must include analytics for “Intrusion, Tripwire, People counting/people gathering, face recognition, Tamper detection”.
PTZ cameras must include “Tripwire, Intrusion, Object Abandon/Missing, Motion Detection, Face Recognition, Tamper detection”.
Automatic number plate recognition (ANPR) cameras, designed to recognise vehicle number plates, must also have edge analytics for “Face Detection”, alongside people counting and gathering, perimeter protection and tamper detection.
Video management software (VMS) must also “support Face Recognition System”. The system “should become part of VMS with perpetual license”.
The documents also state:
“The Manufacturer shall offer a mobile application with the capability to access live video from up to 500 cameras in both iOS and Android/OS platforms.”
The documents specify the following locations for the cameras:
“CCTV cameras are to be strategically placed at suitable height with structure arrangement, so as to identy and capture the face of a particular person roaming around at Critical/vulnerable points.”
For context, a new round of protests began at Jantar Mantar, seeking accountability from the Election Commission of India. The Delhi Police Commissioner de-notified Jantar Mantar as a protest site and designated Ramlila Maidan and two other as a new protest site.
Why it matters: The Indian state continues to use and misuse surveillance technology. Yet, no specific statutory law authorises and regulates the police’s use of facial recognition. For instance, Delhi Police has not disclosed the legal framework or safeguards governing its use of facial recognition.
As MediaNama reported, the standing order that IFF obtained through a Right to Information (RTI) application does not mention facial recognition at all. The Ministry of Home Affairs cited this order in Parliament when asked about safeguards governing facial recognition.
The RFP mentions the word “privacy” only once, in the context of “privacy masking”. This refers to a function that blurs or blacks out selected regions, faces or bodies in live or recorded video footage.
In the context of this RFP, “privacy masking 4 area” might mean that each camera must be capable of defining up to four independent privacy zones within its field of view. Each zone could be selectively blurred, blacked out or pixelated to protect private spaces, such as neighbouring windows and balconies, while the rest of the scene remains visible for surveillance.
We found no references to privacy, the Digital Personal Data Protection Act or the Puttaswamy judgment in the RFP. IFF also noted this absence.
The RFP sets no accuracy standard for facial recognition and does not specify a database against which faces would be matched. It also sets no similarity threshold, rules on who can conduct searches, or retention period for facial data.
These details could be covered by another product, framework or workflow and specified in a separate tender. The RFP is primarily a product requirements document, rather than a legal document that sets limits on how the technology can be used or misused. However, the requirements should still address these issues, given that they will shape the system eventually built or delivered.
If privacy safeguards are not a priority at the product specification stage, how can we expect the resulting products and services to uphold privacy-preserving principles?
Surveillance technology also has real-world consequences. In August 2026, following the July protests at Jantar Mantar, Delhi Police told the Supreme Court in an affidavit that its facial recognition system had identified 2,873 people with criminal antecedents at the protest site between July 20 and 26. Later, The Indian Express found that at least 25 of them were in jail at the time.
Integration with Safe City, which includes predictive policing: The RFP requires the cameras to be “compatible to integrate with any other CCTV cameras installed in other projects of Delhi Police like Safe City Project”. The Delhi Police standing order states that the Safe City system “will have tools for predictive policing”, according to IFF.
Also read:
What remains unanswered from Delhi Police’s RTI reply on live CCTV surveillance & facial recognition
Can The DPDP Act Protect Citizens From CCTV Surveillance?
For You
Read Reasoned by Nikhil Pahwa: How AI is changing our world
Sponsor a MediaNama Event