一名加拿大软件工程师利用人工智能编程工具和价值180美元的摄像头，制作了一套类似美国Flock Safety监控系统的设备，但他的目标并不是追踪普通民众，而是专门识别和记录警车。项目视频发布后，当地警方上门询问其开发目的，并担心相关数据可能暴露警员的日常行踪、工作规律甚至住所位置。这场看似带有讽刺意味的事件，再次引发了人们对自动车牌识别技术、公共监控和个人隐私的讨论。
加拿大软件工程师兼YouTube博主安东尼·西斯蒂利（Anthony Sistilli）近日公开讲述了这段经历。他表示，9月20日，两名来自加拿大安大略省皮尔区地区警察局（Peel Regional Police）的警员来到他家中，询问他为什么要开发一套专门追踪警车的监控设备。
西斯蒂利此前发布了多段视频，展示这套设备的开发过程和实际测试效果。他将自己的项目形容为一套针对Flock监控系统的个人监控设备，其核心理念是利用警方日常使用的技术，反过来观察警察车辆的活动。
这套设备采用自动车牌识别技术，也就是ALPR（Automated License Plate Recognition）。这项技术通常通过摄像头拍摄道路上的车辆，再利用图像识别算法提取车牌号码，并将车牌、拍摄时间和地点等信息保存下来，供后续查询。
与普通的车牌识别设备不同，西斯蒂利设计的系统并不打算记录道路上所有车辆的信息，而是首先识别画面中的车辆是否为警车。只有当系统认为目标车辆属于警用车辆时，才会进一步尝试读取其车牌号码。
为了实现这一功能，他使用了一台价值约180美元的摄像头，并借助人工智能编程工具开发相关软件。整个项目采用相对简单的硬件配置，试图证明，个人开发者不需要昂贵的专业设备，也能够构建具备一定自动识别能力的监控系统。
不过，开发过程并不顺利。最初的版本试图直接处理4K视频，同时完成车辆识别和车牌读取，但系统难以在这种工作负载下有效运行。为此，西斯蒂利降低了视频分辨率，并将识别过程拆分为两个阶段：先判断画面中是否存在警车，再尝试读取其车牌。
这种设计减少了车牌识别软件需要处理的视频数据，也使系统能够在确认车辆类别后再执行后续操作，从而降低计算负担。
尽管如此，这套设备的实际效果仍然相当有限。在一次持续一小时的测试中，系统仅生成了两张警车图像，而且图像中的车牌都难以清晰辨认。换言之，这只是一个能够展示基本技术思路的原型，并不是一套已经具备可靠追踪能力的成熟监控系统。
然而，项目引起的关注远远超过了它目前的技术表现。
西斯蒂利表示，前来询问的警员态度礼貌，但他们担心，如果有人持续收集警车出现的时间和地点，就可能据此推断警员的日常活动规律，甚至利用这些信息对警员进行针对性攻击。
据西斯蒂利转述，警员还提出，若有关监控系统的部署信息和运行数据受到过于宽泛的公开要求约束，可能导致正在进行的调查细节泄露，进而给执法工作带来风险。
西斯蒂利认为，这些担忧恰恰反映出普通民众面对自动车牌识别系统时经常提出的问题：如果持续记录车辆的出现地点和时间，相关数据就可能被用来推断一个人的活动轨迹。
当这种技术用于追踪警车时，警方开始担忧自己的行踪可能受到监控；而当同样的技术被用于追踪普通车辆时，公众也会担心自己的隐私和行动自由受到影响。
这也是该项目最引人关注的地方。西斯蒂利并没有试图入侵警方系统，也没有逆向破解商业监控设备，而是利用普通消费级硬件和现成的人工智能开发工具，搭建了一套功能有限的独立原型。
它所展示的并不是某种只有大型企业或政府机构才能掌握的尖端技术，而是自动识别和记录车辆活动的门槛正在降低。随着摄像头、图像识别模型和数据存储工具越来越容易获得，个人开发者也能够尝试建立类似的系统。
不过，个人原型与商业监控网络之间仍然存在明显差距。西斯蒂利的设备在一小时测试中仅捕捉到两张警车图像，车牌识别效果也不理想。相比之下，商业系统通常具备更成熟的摄像头、图像处理能力、服务器端识别技术和集中式数据库，能够持续处理大量车辆记录。
美国Flock Safety正是这一领域最受关注的企业之一。该公司向执法机构等客户提供自动车牌识别摄像头和相关软件，帮助用户记录经过摄像头的车辆，并在数据库中搜索相关记录。
Flock系统的关键不仅在于拍摄和识别车牌，还在于将不同地点采集的数据连接起来。参与网络的警察部门可以查询来自其他地区摄像头的记录，因此搜索范围不再局限于本辖区安装的设备。
例如，如果警方正在寻找一辆涉嫌参与犯罪活动的汽车，就可以利用车牌号码或车辆特征，在网络中搜索该车辆此前经过哪些监控点，从而帮助调查人员重建其部分行驶轨迹。
但这种能力也带来了隐私方面的争议。批评者认为，如果大量摄像头持续记录普通车辆，并将信息集中到可以跨地区检索的数据库中，那么监控系统就可能从调查特定案件的工具，逐渐演变为覆盖广泛人群的车辆活动追踪网络。
即使系统最初的目的只是寻找被盗车辆或协助调查犯罪，持续积累的历史记录仍可能被用于其他目的。公众通常难以了解哪些机构能够访问这些数据、可以使用什么条件进行搜索，以及相关记录会保存多久。
随着人工智能技术的发展，部分新型系统还能够根据车辆颜色、车型、车身特征和其他视觉线索搜索记录，而不再完全依赖准确的车牌号码。这进一步扩大了相关技术的应用范围，也增加了公众对监控边界的担忧。
西斯蒂利的项目正是在这种背景下引起关注。他试图将自动车牌识别技术的观察方向反转，专门收集警车信息，让执法机构亲身面对同类技术可能带来的隐私问题。
不过，这并不意味着普通民众和警察在所有情况下都面临完全相同的法律和安全风险。警察在执行公务时，车辆活动可能涉及公共安全、执法行动和正在进行的调查；普通民众则可能因为持续监控而暴露私人生活、工作和出行规律。两者在法律地位、职责和具体情境上存在差异。
真正值得讨论的问题，是自动监控系统应当受到怎样的约束，以及相关数据的采集、访问、保存和使用应当遵循哪些明确规则。
对于执法机构而言，车辆监控数据确实能够帮助寻找失窃车辆、追踪犯罪嫌疑车辆和调查案件。但如果缺乏适当的访问控制、审计机制和数据保留限制，系统也可能被滥用，或者在缺乏充分理由的情况下被用于调查普通民众。
对于被监控者而言，最大的困难在于，这些系统往往不会像传统警察拦截那样，让当事人立即意识到自己正在被记录。车辆可能只是经过道路上的一台摄像头，其车牌和相关特征便被自动保存，随后进入一个可以持续查询的数据库。
当大量摄像头相互连接时，单次拍摄看似无关紧要的数据，就可能逐渐形成具有分析价值的长期活动记录。
西斯蒂利的设备目前尚不具备商业系统那样的准确性和规模，但它展示了自动监控技术正在变得更加容易获得。过去需要专门硬件、专业软件和较高开发成本才能实现的功能，如今可以通过消费级设备、开源工具和人工智能编程助手进行初步尝试。
这也意味着，未来有关监控技术的争论可能不再局限于大型企业和政府机构部署了多少摄像头，还会涉及个人开发者能否轻易构建类似系统，以及这些设备在公共空间中的使用应当受到怎样的法律约束。
最终，这起事件并没有证明西斯蒂利已经建立了一套能够有效追踪警车的成熟系统，也没有显示警方认定他的项目违法。它所揭示的是一个更广泛的问题：当自动识别技术能够低成本地记录车辆活动时，人们不仅需要讨论监控能否提高公共安全，也需要认真考虑，谁有权监控谁、数据可以被怎样使用，以及如何防止技术在缺乏监督的情况下侵蚀个人隐私。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
A Canadian software engineer, using an artificial intelligence programming tool and a $180 camera, produced a set of equipment similar to the United States Flock Safety surveillance system, but his goal was not to track ordinary people, but to identify and record police vehicles. After the project video publish, the local police came to inquire about the purpose of its development and feared that the data might reveal the day-to-day whereabouts, working patterns and even the location of the residence of the police officers. This seemingly ironic event rekindled discussions on automatic licence recognition techniques, public surveillance and personal privacy.
Anthony Sistilli, a Canadian software engineer and editor of YouTube, has spoken publicly about this recent experience. He states that, on 20 September, two officers from the District Police Department (Peel Regional Police) in Ontario, Canada, came to his home and asked why he had developed a surveillance device specifically designed to track police vehicles.
Sistili had previously produced publish videos showing the development of the equipment and the practical test results. He described his project as a set of personal surveillance devices for the Flock surveillance system, the core idea being to use the techniques used by the police on a daily basis and, in turn, to observe the activities of police vehicles.
The device is based on an automated license plate recognition technique, known as ALPR. This technology is usually used to photograph vehicles on the road through cameras, then to extract the licence plate numbers using image identification algorithms, and to save information on the licence plates, time and location of the shooting for subsequent reference.
Unlike the normal licence plate recognition device, the system designed by Westerly was not intended to record information on all vehicles on the road, but to first identify the vehicles in the picture as police vehicles. Only if the system considers the target vehicle to be a police vehicle will further attempt to read its licence plate number.
In order to achieve this function, he used a camera valued at approximately $180 and developed relevant software using an artificial intelligence programming tool. The project as a whole uses relatively simple hardware configurations to prove that individual developers do not need expensive professional equipment and are able to build monitoring systems with some automatic recognition capability.
However, the development process was not smooth. The original version sought to process the 4K video directly, while completing vehicle identification and licence plate reading, but the system could not function effectively under this load. To this end, West Timor reduced the video resolution and made the identification process split two stages: Let's see if there's a police car in the picture and then try to read its license plate.
This design reduces the video data to be processed by the licence plate recognition software and also allows the system to perform follow-up operations once a vehicle category has been identified, thus reducing the computational burden.
Nevertheless, the actual effects of the equipment are quite limited. During a one-hour test, the system generated only two police vehicle images, and the plates in the images were difficult to clearly identify. In other words, it is a prototype that can demonstrate basic technical thinking and is not a mature monitoring system that already has a reliable tracking capability.
However, the project generated much more attention than its current technical performance.
Sistili stated that the police officers who came to the interview were polite, but they feared that if there was a continuous collection of the time and place of the presence of the police car, it might be used to infer the routine behaviour of the police officers and even to use that information to target them.
According to Sistili, police officers also suggested that deployment information and operational data on the surveillance system could be compromised by overly broad public requirements, which could lead to the disclosure of details of ongoing investigations, thereby posing a risk to law enforcement.
According to Sistili, these concerns reflect the problems that are often raised by ordinary people in the face of the automated license recognition system: If vehicles are continuously recorded where and when they occur, the data may be used to extrapolate a person ' s movement trajectory.
When such technology is used to track police vehicles, the police begin to fear that their movements may be monitored, and when the same technology is used to track vehicles, the public also fears that their privacy and freedom of movement may be affected.
This is also the most interesting place for the project. Instead of attempting to hack the police system and reverse the deciphering of commercial surveillance equipment, Westerly built a stand-alone prototype with limited functionality using ordinary consumer hardware and readily available artificial intelligence development tools.
Rather than demonstrating some sophisticated technology that can only be captured by large enterprises or government agencies, the threshold for automatic identification and recording of vehicle activity is being lowered. As cameras, image identification models and data storage tools become more readily available, individual developers can try to establish similar systems.
However, there is still a clear gap between individual prototypes and commercial monitoring networks. Westerly's equipment captured only two police vehicle images in an hour's test, and licence plate recognition was not optimal. In contrast, commercial systems usually have more sophisticated cameras, image processing capabilities, server-end identification techniques and centralized databases that enable the continuous processing of large volumes of vehicle records.
Flock Safety of the United States is one of the most concerned businesses in this area. The company provides automatic licence plate recognition cameras and related software to clients, including law enforcement agencies, to help users record vehicles that have passed the camera and to search the relevant records in the database.
The key to the Flock system is not only to photograph and identify licence plates, but also to connect data collected at different locations. Police units participating in the network have access to records from other regional cameras, so the search is no longer limited to equipment installed in their jurisdictions.
For example, if the police are looking for a vehicle suspected of involvement in criminal activities, the vehicle's licence plate number or vehicle signature can be used to search the network for points of surveillance that the vehicle had previously passed, thus helping investigators to re-establish some of their tracks.
However, this capacity also leads to privacy disputes. Critics believe that if a large number of cameras continue to record ordinary vehicles and centralize information in databases that can be retrieved across regions, the surveillance system may evolve from a case-specific investigative tool to a vehicle tracking network covering a wide range of people.
Even if the original purpose of the system was simply to locate stolen vehicles or assist in the investigation of crimes, the continuing accumulation of historical records could be used for other purposes. It is often difficult for the public to know which agencies are able to access the data, what conditions are available for searching and how long the relevant records will be kept.
With the development of artificial intelligence technology, some of the new systems will also be able to search for records based on vehicle colours, model, body characteristics and other visual clues, rather than relying entirely on accurate licence plate numbers. This has further expanded the application of relevant technologies and increased public concern about monitoring borders.
It is in this context that the project in West Timor has attracted attention. He tried to reverse the observation direction of the automatic vehicle recognition technology by gathering police vehicle information and by exposing law enforcement agencies to privacy problems that might arise from similar technologies.
This does not mean, however, that the general population and the police face exactly the same legal and security risks in all circumstances. In the exercise of their official duties, the police may be involved in vehicle activities related to public safety, law enforcement operations and ongoing investigations, while the general population may be exposed to private life, work and travel patterns as a result of continuous surveillance. There are differences in their legal status, responsibilities and circumstances.
What is really worth discussing is how automated surveillance systems should be regulated and what clear rules should govern the collection, access, preservation and use of relevant data.
For law enforcement agencies, vehicle surveillance data do help to locate stolen vehicles, track suspected vehicles and investigate cases. However, in the absence of adequate access controls, audit mechanisms and data retention restrictions, systems may also be misused or used to investigate the general population without sufficient justification.
The greatest difficulty for those under surveillance is that these systems often do not make them immediately aware that they are being recorded, as traditional police intercepts do. A vehicle may be simply a camera passing through the road, its licence plates and associated features are automatically preserved and subsequently entered into a database that is continuously searchable.
When a large number of cameras are interconnected, a single shot of seemingly irrelevant data may gradually result in a record of long-term activities of analytical value.
Westerly's equipment is not yet as accurate and as large as the commercial system, but it shows that automated surveillance technology is becoming more readily available. The functions that had previously required specialized hardware, specialized software and higher development costs could now be piloted through consumer-level equipment, open-source tools and artificial intelligent programming assistants.
This also means that future debates on surveillance techniques may no longer be limited to the number of cameras deployed by large enterprises and government agencies, but may also involve the ease with which individual developers can construct similar systems and the legal constraints that should be placed on the use of such equipment in public space.
Ultimately, the incident did not prove that Westerly had a mature system in place to effectively track the police car, nor did it show that the police had found his project to be illegal. It reveals a broader problem: when AITs are able to record vehicle activities at low cost, people need to discuss not only whether surveillance can improve public safety, but also carefully, who has the right to monitor who, how data can be used, and how to prevent technology from eroding personal privacy in the absence of supervision.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.