A privacy protection type intelligent camera system

Abstract

This invention discloses a privacy-preserving smart camera system, comprising: an image sensor for acquiring raw image signals of a scene; a target recognition module, based on a deep learning model, for real-time detection of faces, bodies, or user-specified specific object regions in the scene; a privacy region processing module, for performing at least one privacy protection operation among blurring, pixelation, or occlusion on non-target regions detected by the target recognition module; and a data encryption module, employing an end-to-end encryption algorithm to locally encrypt the privacy-preserving video stream, ensuring data transmission and storage security. This invention, through real-time target recognition at the device side, retains only the target region image needed by the user, while non-target regions undergo irreversible processing using various adjustable privacy methods, eliminating the risk of redundant acquisition and leakage of sensitive information from the source, and solving the problem of uncontrollable image acquisition range in existing cameras.

Description

Technical Field

[0001] This invention relates to the field of smart camera technology, and specifically to a privacy-protecting smart camera system that combines monitoring functions with privacy protection capabilities. Background Technology

[0002] With the rapid development of IoT technology and the smart home industry, smart cameras have been widely penetrated into various fields such as homes, offices, transportation, and retail, becoming core devices for security monitoring, identity recognition, and remote interaction. According to industry statistics, the global shipment volume of smart home cameras has grown at an average annual rate of over 30% in recent years, with its application scenarios continuously expanding and functional requirements constantly upgrading.

[0003] However, existing smart camera systems have significant technical shortcomings in practical applications, mainly in the following aspects: Privacy risks are prominent: Traditional smart cameras use a "full-screen capture - cloud storage / transmission" model. The captured video stream contains all information within the scene, including not only the target information the user needs (such as faces and bodies), but also sensitive privacy information such as home furnishings, office documents, and personal behavior. If this raw data is compromised during transmission or storage due to security incidents such as hacker attacks or platform data breaches, user privacy will be illegally obtained, leading to serious security risks. For example, in 2023, a well-known surveillance platform experienced a data breach, resulting in the illegal dissemination of home surveillance videos from millions of users, causing a severe negative social impact.

[0004] Uncontrollable image capture range: Even if existing cameras only need to perform face or body detection functions, they will still indiscriminately capture images of the entire monitoring scene, resulting in the redundant collection of a large amount of irrelevant and sensitive information. For example, in a video conferencing scenario, users only need to show their own face area, but the camera will simultaneously capture information such as the layout of the meeting room, irrelevant actions of other people, and desktop files, which may lead to the leakage of trade secrets or personal privacy.

[0005] Reliance on cloud processing leads to insufficient security and real-time performance: To reduce on-device hardware costs, some camera products rely on cloud servers for core functions such as target recognition and privacy processing. In this model, raw images must be transmitted to the cloud, which not only poses a risk of data leakage during transmission but also causes processing delays due to network latency, failing to meet the needs of real-time monitoring and instant identity verification scenarios. Especially in environments with unstable network signals, issues such as image stuttering and recognition delays occur, severely impacting the user experience.

[0006] Insufficient flexibility in privacy protection strategies: Some existing cameras with basic privacy protection functions only support fixed modes of privacy processing (such as blurring fixed areas), and cannot customize the target area or adjust the privacy protection intensity according to user needs, making it difficult to adapt to personalized privacy protection needs in different scenarios.

[0007] Therefore, how to achieve efficient and accurate target recognition at the edge, provide flexible and reliable privacy protection for non-target areas, and ensure the security of data transmission and storage has become a key issue that urgently needs to be addressed in the current field of smart camera technology. This invention addresses the shortcomings of the prior art by proposing a privacy-protecting smart camera system to fill this gap in the technology. Summary of the Invention

[0008] The core objective of this invention is to overcome the technical shortcomings of existing smart cameras, such as high privacy leakage risks, uncontrollable image capture range, reliance on cloud processing, and rigid privacy protection strategies, and to provide a privacy-protecting smart camera system. This system can perform real-time target recognition and privacy processing of non-target areas on the device side, achieving a deep balance between monitoring functionality and privacy security. It also features flexible privacy protection strategies and reliable data security guarantees, adapting to the needs of various application scenarios.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a privacy-protecting smart camera system, comprising: Image sensors are used to acquire raw image signals of a scene; The target recognition module, based on a deep learning model, detects faces, bodies, or specific object regions specified by the user in a scene in real time. The privacy region processing module performs at least one privacy protection operation among blurring, pixelation, or occlusion on the non-target region detected by the target recognition module. The data encryption module uses an end-to-end encryption algorithm to locally encrypt the video stream after privacy protection processing, ensuring the security of data transmission and storage; The image output module outputs the privacy-protected and encrypted video stream for user viewing or for integration with higher-level systems.

[0010] Preferably, the deep learning model used in the target recognition module is a lightweight target detection model, which supports real-time inference on the edge device, with a detection latency of ≤50ms and a face / body detection accuracy of ≥95%.

[0011] Preferably, the privacy area processing module supports user-defined privacy processing strategies, allowing users to select a single privacy protection operation or a combination of operations, and the blurring degree and pixel block size are adjustable.

[0012] Preferably, the data encryption module uses the AES-256 encryption algorithm to achieve end-to-end encryption, and also supports hierarchical access control, allowing only authorized users to decrypt and view the video stream.

[0013] Preferably, the image sensor is a high-definition CMOS sensor with ≥10 million pixels, a frame rate ≥25fps, and supports a recognition distance of 0.3-5m, adapting to the image acquisition needs of various indoor and outdoor scenarios.

[0014] Preferably, the target recognition module allows users to input specific object features through a terminal APP, enabling accurate detection and recognition of custom objects.

[0015] Preferably, the system completes image acquisition, target recognition, privacy processing, and data encryption entirely on the device side, without uploading the original unprocessed images to the cloud.

[0016] A privacy-protecting smart camera system is used in home monitoring, smart door locks, office monitoring or video conferencing scenarios. It can automatically match the corresponding target recognition rules and privacy protection strategies according to the needs of the scenario.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. More thorough and accurate privacy protection: This invention uses real-time target recognition on the device side to retain only the target area image that the user needs. Non-target areas are processed irreversibly using a variety of adjustable privacy processing methods, eliminating the risk of redundant collection and leakage of sensitive information from the source and solving the problem of uncontrollable image collection range of existing cameras. 2. Localized processing ensures security and real-time performance: Core target recognition, privacy processing, and data encryption functions are all completed on the device side, eliminating the need to upload the original images to the cloud. This avoids the risk of data leakage during transmission and eliminates network latency in cloud processing, ensuring the real-time performance of the monitoring screen and meeting the needs of scenarios requiring instant identity recognition and real-time monitoring. 3. Highly flexible privacy policy: It supports users to customize target types, privacy processing methods and processing strengths, and can also preset privacy policies for multiple scenarios to adapt to the personalized needs of different application scenarios such as home monitoring, office monitoring, video conferencing, and smart door locks, thus solving the shortcomings of rigid privacy protection policies in existing products; 4. Reliable data security: It adopts the AES-256 end-to-end encryption algorithm and hardware security chip to store keys, combined with hierarchical access control, to build a full-process data security barrier of "collection-processing-transmission-storage", effectively resisting security threats such as data theft and tampering; 5. Wide applicability and strong practicality: The hardware configuration is adapted to multiple indoor and outdoor scenarios, and the software functions support the monitoring and privacy protection needs of multiple fields such as home, office, retail, and transportation. It does not require major modifications for specific scenarios and has broad application prospects and market value. Attached Figure Description

[0018] Figure 1 This is a block diagram of a privacy-protecting smart camera system according to the present invention; Figure 2 This is a flowchart of a privacy-protecting smart camera system according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figures 1 to 2 The present invention provides a technical solution: System components: The privacy-protecting smart camera system of this invention comprises five core modules, which work together to achieve end-to-end privacy protection and security control from image acquisition to video output: Image sensor: It adopts a high-definition CMOS sensor with ≥10 million pixels and a frame rate of ≥25fps. It supports autofocus and light adaptive adjustment, and the recognition distance covers 0.3-5m. It can acquire clear and stable raw image signals under different indoor and outdoor lighting conditions, providing high-quality data input for subsequent target recognition.

[0021] Target Recognition Module: Built upon lightweight deep learning target detection models (such as YOLOv5-tiny and MobileNet series optimized models), this module is deployed on edge devices. It features real-time detection capabilities, accurately recognizing faces and bodies in a scene. It also allows users to input feature information of specific objects (such as packages, documents, or specific items) via a terminal app, enabling the recognition of custom targets. The module has a detection latency of ≤50ms, a face / body detection accuracy of ≥95%, and a custom target recognition accuracy of ≥90%, ensuring rapid target area localization without cloud reliance.

[0022] Privacy Area Processing Module: This module works in conjunction with the target recognition module to perform privacy protection operations on non-target areas identified after target recognition. It supports three core processing methods: blurring (Gaussian blur, blur intensity adjustable via the user app, σ value range 1-5), pixelation (pixel block size adjustable from 8x8 to 32x32), and occlusion processing (black occlusion block or custom watermark occlusion). Users can choose a single processing method or a combination of processing methods according to scenario requirements, or preset processing strategies for different scenarios, which the system will automatically match and execute.

[0023] Data encryption module: Employs the AES-256 end-to-end encryption algorithm to locally encrypt the privacy-protected video stream. The encryption process is completed on the device side; the key is set by the user via the terminal app and stored locally on a secure chip, supporting periodic key updates. Simultaneously, the module supports tiered access control, allowing users to set different permission levels such as administrator and visitor. Only authorized users can decrypt and view the video stream using the key, ensuring data security during transmission (e.g., remote preview) and storage (e.g., local SD card storage), preventing unauthorized tampering or theft.

[0024] Image output module: Supports multiple output methods, including local HDMI interface display, remote preview via terminal APP, and integration with smart home platforms (such as Mi Home and Huawei HarmonyOS) or office systems (such as video conferencing software). The output video stream adopts the H.265 encoding format, balancing image quality and transmission efficiency to ensure users receive a clear and smooth viewing experience in different usage scenarios, with an output latency of ≤100ms.

[0025] The workflow of the privacy-protecting smart camera system of this invention is as follows: Step 1: Image Acquisition. The image sensor acquires raw image signals of the monitored scene in real time, and automatically adjusts parameters such as sensitivity and white balance according to the ambient light intensity to ensure that the acquired images are clear and usable; Step 2: Target Recognition. The raw image signal is transmitted to the target recognition module. The module uses a deployed lightweight deep learning model to detect target regions (faces, bodies, or user-defined objects) in the image in real time and outputs the coordinates and range information of the target regions. Step 3: Privacy Protection Processing. Based on the target area information output by the target recognition module, the privacy area processing module determines the non-target area range and performs irreversible privacy protection operations on the non-target area according to the user-preset privacy processing strategy (such as blurring and pixelation), retaining only the original clear image of the target area; Step 4: Data Encryption. The data encryption module performs local AES-256 encryption on the privacy-protected video stream to generate encrypted video data, while simultaneously recording access permission information; Step 5: Image Output. The image output module outputs the encrypted video stream according to the user-selected output method (local display, remote preview, system integration). Authorized users can view the processed video after decrypting it with a key.

[0026] Example 1: Home surveillance application System configuration: Employs a 12-megapixel CMOS image sensor, 30fps frame rate, and recognition distance of 0.5-5m; The target recognition module deploys the YOLOv5-tiny optimized model, supporting face and human detection; The privacy processing strategy is set to "clearly retain faces, and Gaussian blur the rest of the area (σ=3)"; The data encryption module uses AES-256 encryption, and users can set access keys through a mobile APP.

[0027] Operating Process: The camera is installed in the living room, and the image sensor captures the living room scene in real time. The target recognition module detects faces / body areas in the scene in real time, accurately locating the facial positions of family members or visitors. The privacy area processing module performs Gaussian blur processing on non-target areas in the living room, such as sofas, TVs, desktop files, and decorative items. The data encryption module encrypts the processed video stream locally. Part of the encrypted video data is stored on a local 1TB SD card, and the other part is transmitted to the user's mobile APP via Wi-Fi. The user can enter a preset key through the mobile APP to decrypt and view the monitoring screen containing only the clear areas of faces, effectively protecting the privacy of the home environment and personal behavior, while not affecting the monitoring of the safety of family members.

[0028] Application results: Target recognition latency ≤30ms, face detection accuracy 98%, non-target areas cannot be restored to their original information after blurring, no data leakage occurred during video stream transmission, and no security vulnerabilities were found in locally stored data after brute-force attack testing.

[0029] Example 2: Video Conferencing Scenario Application System configuration: Employs a 10-megapixel wide-angle CMOS image sensor with a 120° field of view and a frame rate of 25fps; the target recognition module supports face and upper body region detection, and also allows users to customize "meeting documents" as specific recognition objects; the privacy processing policy is set to "clearly retain the participant's face and upper body, replace the background area with a virtual background, and pixelate the meeting document area (16x16 pixel block)"; the data encryption module interfaces with video conferencing software (such as Tencent Meeting and Zoom) and supports permission synchronization.

[0030] Working process: Cameras are deployed in the office conference room. After the meeting begins, the image sensor captures a panoramic view of the conference room; the target recognition module quickly detects the faces and upper bodies of the participants, and simultaneously identifies the meeting document area on the desktop; the privacy area processing module replaces the background areas such as the conference room wall decorations, other irrelevant personnel, and the scene outside the window with the system's built-in virtual office scene background, and performs pixelation processing on the desktop meeting document area; the encrypted video stream is transmitted to the video conferencing software, and participants can view the processed image after verifying their meeting permissions. They can only clearly see the speaker's face and upper body, but cannot obtain details of the conference room layout or the content of the meeting documents.

[0031] Application results: Virtual background replacement is seamless with no screen tearing; pixelated files cannot recognize text content; target area tracking accuracy is 96%; it is suitable for multi-person switching speaking scenarios and meets the privacy protection and information security needs of remote work.

[0032] Example 3: Smart door lock application scenario System configuration: It adopts an 8-megapixel low-power CMOS image sensor with standby power consumption ≤50mW and a recognition distance of 0.3-1.5m; the target recognition module is optimized to a low-power version, which only wakes up when a person is detected approaching, and supports face detection; the privacy processing strategy is set to "clearly retain the visitor's face, and handle the obstruction of the door environment and the neighbor's door area"; the data encryption module is linked with the smart door lock main control chip, and the key is stored in the door lock security chip.

[0033] Working process: When a visitor approaches the smart door lock (distance ≤ 1.5m), the image sensor is activated and captures images of the visitor and the doorway area; the target recognition module quickly detects the visitor's face area and outputs the face coordinates; the privacy area processing module performs blackout processing on non-target areas such as the shoe cabinet, doorbell, neighbor's doorplate, and corridor environment at the door; the encrypted face image is transmitted to the door lock's main control chip and simultaneously pushed to the user's mobile APP; the user can view the clear image of the visitor's face through the APP, confirm their identity, and remotely authorize the door to open, thus avoiding privacy leaks in the doorway environment.

[0034] Application results: The wake-up delay when people approach is ≤300ms, the face detection accuracy is 97%, the non-target area is completely blocked, and the door lock has a battery life of ≥6 months in low power mode, balancing security and practicality.

Claims

1. A privacy-protecting smart camera system, characterized in that, include: Image sensors are used to acquire raw image signals of a scene; The target recognition module, based on a deep learning model, detects faces, bodies, or specific object regions specified by the user in a scene in real time. The privacy region processing module performs at least one privacy protection operation among blurring, pixelation, or occlusion on the non-target region detected by the target recognition module. The data encryption module uses an end-to-end encryption algorithm to locally encrypt the video stream after privacy protection processing, ensuring the security of data transmission and storage; The image output module outputs the privacy-protected and encrypted video stream for user viewing or for integration with higher-level systems.

2. The privacy-protecting smart camera system according to claim 1, characterized in that, The target recognition module uses a lightweight target detection model, which supports real-time inference on the edge device, with a detection latency of ≤50ms and a face / body detection accuracy of ≥95%.

3. The privacy-protecting smart camera system according to claim 1, characterized in that, The privacy area processing module supports user-defined privacy processing strategies, allowing users to select a single privacy protection operation or a combination of operations, and the blurring degree and pixel block size are adjustable.

4. The privacy-protecting smart camera system according to claim 1, characterized in that, The data encryption module uses the AES-256 encryption algorithm to achieve end-to-end encryption, and also supports hierarchical access control, allowing only authorized users to decrypt and view the video stream.

5. The privacy-protecting smart camera system according to claim 1, characterized in that, The image sensor is a high-definition CMOS sensor with ≥10 million pixels, a frame rate ≥25fps, and supports a recognition distance of 0.3-5m, adapting to image acquisition needs in various indoor and outdoor scenarios.

6. A privacy-protecting smart camera system according to any one of claims 1-5, characterized in that, The target recognition module allows users to input specific object features through a terminal APP, enabling accurate detection and recognition of custom objects.

7. A privacy-protecting smart camera system according to any one of claims 1-5, characterized in that, The system completes image acquisition, target recognition, privacy processing, and data encryption entirely on the device side, without uploading raw, unprocessed images to the cloud.

8. A privacy-protecting smart camera system according to any one of claims 1-5, which is applied to home monitoring, smart door locks, office monitoring or video conferencing scenarios, and can automatically match the corresponding target recognition rules and privacy protection strategies according to the needs of the scenario.

Images