A vehicle-mounted seamless entry system based on gait and microphysiological characteristics

Abstract

This invention discloses a vehicle-mounted contactless entry system based on gait and microphysiological features, comprising: a contactless human feature perception module (without a camera) that collects gait, body shape, and microphysiological features of a person; a local AI feature modeling and comparison module that establishes a whitelist of family members, generates temporary profiles for strangers and automatically clears them; and a contactless entry control module that automatically unlocks the vehicle from a distance of 3 to 5 meters, with the user acting as the key. The perception module employs non-contact sensing technologies, including but not limited to millimeter-wave radar, ultrasound, UWB, and Wi-Fi. Microphysiological features include breathing, heartbeat, and micro-vibrations of the limbs. Gait models at various speeds are collected during whitelist registration, supporting speed extrapolation, identifying rapidly running family members, and possessing stranger identification, video recording alarms, and gait self-learning functions. This invention achieves contactless entry by actively sensing the vehicle and requiring no user intervention, balancing privacy protection and ease of use.

Description

Technical Field

[0001] This invention relates to the fields of intelligent connected vehicles, in-vehicle contactless perception, AI biometric recognition, and keyless vehicle entry control technology. Specifically, it relates to a system and method for actively collecting human gait and micro-physiological characteristics through a non-contact sensing module to achieve identity recognition and contactless unlocking. Background Technology

[0002] Existing vehicle access control solutions can be broadly categorized as follows: The first category relies on physical media carried by the user, including traditional mechanical keys, remote keys, mobile phone Bluetooth, NFC cards, or smart bracelets. These solutions require users to carry a specific device; if the device is forgotten, runs out of power, or the signal is interfered with, the vehicle cannot be unlocked. In daily life, users frequently encounter situations where they forget their keys, their phones are out of power, or Bluetooth connections fail, rendering the vehicle unusable and severely impacting the user experience. Even with the device, users still need to actively take it out or bring it near a specific sensing area, which remains inconvenient. The second category relies on user-initiated physical actions, such as kicking the tailgate, waving, or touching the door handle. While this eliminates the need for keys, it requires users to consciously perform specific actions and cannot distinguish between different users. Anyone performing the correct action can open the door, posing a security risk and failing to provide personalized access control for family members. The third category relies on biometric recognition, including facial recognition, fingerprint recognition, and voiceprint recognition. Facial recognition requires high-definition cameras, which are susceptible to changes in lighting, rain, snow, obstructions, and nighttime conditions, resulting in poor recognition stability. Furthermore, it involves facial image collection, posing a privacy risk and requiring compliance with increasingly stringent personal information protection regulations. Fingerprint and voiceprint recognition solutions require users to actively contact or speak, which is also not seamless. The common inherent flaw of these existing technologies is that users must actively cooperate with the vehicle (carrying devices, making movements, and cooperating with biometric data collection), while the vehicle remains in a passive response position. If a user forgets their keys, their phone is out of battery, their hands are occupied, or they are in a state of inconvenience, they cannot get into the vehicle. Therefore, there is an urgent need for a seamless entry solution where the vehicle actively senses the user's presence, requiring no user interaction, and using the user's own biometrics as the sole authentication key, completely eliminating reliance on external devices. Summary of the Invention

[0003] This invention addresses the shortcomings of existing technologies that require users to actively cooperate with the vehicle by providing a vehicle-mounted contactless entry system based on gait and microphysiological characteristics. This invention revolutionizes the traditional model: the vehicle actively collects the gait, body contour, and microphysiological characteristics (breathing, heartbeat, limb micro-vibrations, etc.) of approaching individuals through a contactless human feature sensing module. Using the user's gait and microphysiological characteristics as a unique digital key, it automatically performs identity recognition and authentication, achieving true door opening and contactless entry. Contactless entry is set as a core independent function, while security and video recording alarms are optional additional functions. Temporary identity files are managed for unfamiliar personnel, enabling cross-identity association through feature comparison and automatic clearing of expired files. A gait feature self-learning update mechanism is added to adapt to natural changes in human body posture and maintain long-term recognition accuracy. Technical solution

[0004] A vehicle-mounted non-intrusive entry system based on gait and microphysiological features includes: a non-intrusive human feature perception module, a local AI feature modeling and comparison module, a non-intrusive entry control module, and optional intelligent stranger behavior identification module and intelligent video recording linkage alarm module.

[0005] Non-sensory human feature perception module The non-contact human feature perception module does not rely on any facial recognition cameras. Its function is to collect the gait, body contours, and micro-physiological characteristics of people around the vehicle. This module employs non-contact sensing technology, including but not limited to one or more of millimeter-wave radar, ultrasonic sensors, UWB radar, and Wi-Fi sensing modules. This module allows family members to unlock the vehicle without carrying any physical keys or electronic devices; their own biometrics become the sole credential for unlocking the vehicle.

[0006] In a preferred embodiment, the sensing module employs millimeter-wave radar (operating at a frequency of 60 GHz or 77 GHz). Millimeter-wave radar extracts the target's distance, velocity, angle, and micro-Doppler features by transmitting frequency-modulated continuous waves and receiving echoes reflected from the human body.

[0007] For gait feature extraction: The radar transmits multiple linear frequency modulated pulses within a coherent processing interval, performs a range-Doppler two-dimensional Fourier transform on the echoes, and obtains the range-Doppler spectrum; when a person walks, the swinging of the limbs will generate micro-Doppler modulation, and the time-frequency spectrum is obtained through short-time Fourier transform. From this, parameters such as gait period, single step duration, swing amplitude, and alternation frequency of the two legs are extracted to form a gait feature vector.

[0008] For body contour extraction: using multiple input multiple output virtual array technology, the azimuth and pitch angles of the target are obtained through direction of arrival estimation. Combined with distance information, a 3D point cloud is generated. After clustering and filtering, the external dimensions (height, shoulder width, etc.) of the target are obtained.

[0009] For microphysiological feature extraction: the human chest cavity undergoes periodic micro-displacements due to respiration and heartbeat (respiratory amplitude approximately 5-12 mm, heartbeat amplitude approximately 0.2-0.5 mm); radar echo phase is sensitive to sub-millimeter-level displacement. Phase change sequences are obtained through phase dewinding, and then bandpass filtering (respiration: 0.1-0.5 Hz, heartbeat: 0.8-2.0 Hz) separates the respiration and heartbeat waveforms. Fourier transform is then used to extract the frequency peaks to obtain the respiratory rate and heart rate. The system employs adaptive clutter suppression technology to filter out static background interference and random noise, ensuring stable extraction of vital signs signals from targets in complex environments.

[0010] As another preferred embodiment, the sensing module can employ an ultrasonic sensor, utilizing the Doppler effect to detect ultrasonic frequency shifts generated by human movement and extract gait and limb micro-vibration features. Ultrasonic sensors are low-cost, consume little power, and have good penetration through plastics.

[0011] As another preferred implementation, the sensing module can employ UWB (ultra-wideband) radar, which can obtain centimeter-level distance resolution by emitting nanosecond-level narrow pulse signals, enabling precise measurement of micro-displacement of the human chest cavity (breathing, heartbeat), while the high temporal resolution is beneficial for the extraction of gait micro-Doppler features.

[0012] As another preferred implementation, the sensing module can employ Wi-Fi sensing technology. By analyzing changes in channel state information (CSI) along the propagation path of the Wi-Fi signal, it extracts the signal amplitude and phase fluctuations caused by human movement, thereby identifying gait and breathing patterns. This approach can reuse the vehicle's existing onboard Wi-Fi module without incurring additional hardware costs.

[0013] As a more preferred embodiment, the sensing module can be a fusion sensing module of two or more of the above technologies, such as the fusion of millimeter-wave radar and UWB radar. By utilizing the advantages of millimeter-wave radar in long-range gait recognition and UWB radar in short-range high-precision micro-motion detection, data fusion is performed through Kalman filtering or neural networks to further improve the recognition accuracy and environmental robustness.

[0014] This module can be concealed and installed within the plastic exterior or interior cavities of a vehicle (such as the inside of the bumper, rearview mirror housing, or the interior trim panel along the lower edge of the door). It requires no cutting of the vehicle's sheet metal, no drilling, and does not alter the vehicle's structural safety framework. The plastic components cause minimal attenuation of millimeter-wave signals, allowing it to penetrate the plastic and collect signals normally. It does not collect, analyze, or store any facial image information throughout the entire process.

[0015] Feature extraction and matching strategies under dynamic motion conditions The system continuously tracks the target's motion state using millimeter-wave radar, including stationary motion, walking, rapid movement (such as jogging or brisk walking), speed changes, and turning. The radar's Doppler velocity measurement and range-time trajectory analysis enable real-time determination of the target's instantaneous velocity, acceleration, and changes in direction of motion.

[0016] When a target approaches the vehicle at different speeds, the system employs the following adaptive strategy: (1) Velocity-adaptive feature extraction: For fast-moving targets, the radar's micro-Doppler spectrum will show higher frequency broadening and denser time-frequency changes. The system dynamically adjusts the length of the time-frequency analysis window to ensure that gait features can still be completely captured under high-speed movement.

[0017] (2) Dynamic time warping comparison: The real-time acquired gait waveform is dynamically warped and aligned with the reference gait waveform registered in the whitelist to eliminate the time axis stretching caused by speed differences.

[0018] (3) Multi-level threshold and confidence fusion: When the target is in a high-speed motion state, the matching threshold for micro-physiological features such as breathing / heartbeat is appropriately relaxed, while the matching weight of gait and body shape is increased; when the target decelerates or stops, high-precision verification is immediately restored.

[0019] For family members approaching at a speed higher than the maximum registered speed (such as running at full speed), the system can still achieve effective matching through a speed extrapolation model and dynamic time warping algorithm, without needing to collect corresponding speed samples during registration, while strictly filtering out false triggers from non-whitelisted personnel.

[0020] Local AI Feature Modeling and Comparison Module The local AI feature modeling and comparison module is deployed in the vehicle domain controller or edge computing unit.

[0021] Whitelist creation and feature entry process: During initial registration, family members need to complete a set of preset actions around the vehicle. The system collects multi-dimensional features to establish a complete biometric template library. Specifically, this includes: (1) Natural gait acquisition: Walk back and forth 1-2 times at normal speed and extract the baseline gait feature vector.

[0022] (2) Dynamic gait acquisition: Walk 1-2 round trips each at brisk walking (approximately 1.5-2 m / s) and jogging (approximately 2.5-3.5 m / s) respectively, extract speed adaptive parameters, and construct a dynamic gait model. This model supports speed extrapolation and can identify family members running at a speed higher than the registered speed.

[0023] (3) Body contour collection: Standing still, extract external features such as height and shoulder width.

[0024] (4) Acquisition of standing micro-motion characteristics: Stand still for about 10 seconds and collect respiratory waveforms, micro-tremor spectrum, heartbeat waveforms and center of gravity shift trajectory.

[0025] (5) Approaching micro-motion samples: record the distance-time curve and radar feature change pattern of walking towards the car door and raising the hand.

[0026] After normalizing the above five types of features, the system merges them into a fixed-dimensional comprehensive feature vector (e.g., 512-dimensional). This vector is then used to generate an irreversible, unique digital identity code via hash mapping and stored in the vehicle-mounted encrypted whitelist database. The whitelist supports multiple family members, each of whom can store an independent comprehensive feature template, including a natural gait template, a dynamic gait model, and microphysiological feature templates.

[0027] During real-time comparison, feature vectors close to the target are extracted online and their similarity (using cosine or Euclidean distance) is calculated with all feature templates in the whitelist. A dual threshold is set: a low threshold of 0.65 for suspected matches and a high threshold of 0.85 for confirmed matches. Simultaneously, respiratory / heartbeat waveforms are used for liveness detection to exclude video playback or spoofing attacks. For rapidly moving targets, a pre-stored dynamic gait model from the family member whitelist is used for speed-normalized matching. Only when the gait similarity exceeds the high threshold and the microphysiological signals match liveness characteristics is the target identified as a registered family member.

[0028] Temporary identity file management for strangers For unfamiliar individuals not on the whitelist, the system does not perform an unlocking operation, but generates a temporary identity profile for behavioral statistics and suspicious identification. When a stranger enters the sensing area for the first time, its feature vector is extracted, a temporary ID is assigned, the first appearance time is recorded, the initial appearance count is set to 1, and its movement trajectory is saved.

[0029] When the same stranger leaves and re-enters, the system extracts the current features and compares them with all cleared temporary files. If the feature similarity with a file exceeds a preset threshold (e.g., 0.8), it is determined to be the same person, and the file's appearance count, most recent appearance time, and trajectory are updated; otherwise, a new temporary file is created.

[0030] Temporary files are retained for 24 hours. Files of personnel that have not reappeared after 24 hours are automatically deleted. If the number of files exceeds a preset limit (e.g., 100), the files that have not been updated for the longest time are deleted first. All feature comparisons and identity determinations are completed locally on-board edge computing, and feature encoding data is stored locally only and is not uploaded to the cloud.

[0031] Seamless entry control module The contactless entry control module connects with the vehicle's original door lock electronic control system via CAN bus or Ethernet. When the comparison result indicates a registered family member is within the preset unlocking distance (typically 3-5 meters), the control module generates an unlocking command, driving the door lock actuator to automatically unlock the door, and optionally sends a welcome signal (lights, rearview mirrors unfolding, etc.). Family members can be identified from any direction, without needing to walk to a specific location or perform any active operation.

[0032] Optional extended function modules Stranger Behavior Intelligent Identification Module (Optional Configuration): Based on the number of times the same stranger appears and their movement trajectory recorded in the temporary file, the module counts the frequency of appearances within a sliding time window (e.g., 1 hour) and analyzes whether the trajectory contains abnormal patterns such as walking around the vehicle, pacing back and forth, or staying for more than 2 minutes at a time. When the "number of appearances within the window ≥ 2 times" and the "trajectory is abnormal" are met, it is judged as suspicious reconnaissance behavior. The system transfers the file to a local suspicious behavior storage device (long-term storage, not automatically deleted) for subsequent alarms or record queries.

[0033] Intelligent video recording and alarm module (optional configuration): When suspicious behavior is detected, the vehicle dashcam and in-vehicle monitoring camera are automatically activated to record and store the video. At the same time, the vehicle lights flash and the horn sounds intermittently. Alarm information (including time, location, short video clips or pictures) is pushed to the owner's mobile APP through the vehicle communication module.

[0034] Gait feature self-learning update mechanism The system incorporates a self-learning update module. Since family members' body postures change slowly due to age, weight, and other factors, the recognition rate of a fixed gait template may decrease after long-term use. Upon each successful unlock, the system updates the extracted gait features by incremental averaging or Kalman filtering with existing features in the whitelist: New feature = (1-α) × Old feature + α × Current feature, where α is the learning rate (typically 0.1). To prevent outliers from contaminating the template, a confidence counter is used. An update is only performed when there are multiple consecutive successful recognitions (e.g., 5 times) and the feature change trend is consistent. If the recognition confidence remains low for more than 10 consecutive times, the system can prompt the user for a second registration verification to ensure the whitelist always reflects the user's latest status.

[0035] Beneficial technical effects 1. This invention uses a non-contact human feature perception module, which does not rely on a face camera, completely eliminating the risk of privacy collection, and is not affected by light, rain, snow, or day and night, and can work stably in all weather conditions.

[0036] 2. It integrates gait and microphysiological biometrics for identification, making it highly unique and difficult to counterfeit. Breathing / heartbeat liveness detection further enhances security.

[0037] 3. Enables seamless unlocking from a distance of 3 to 5 meters. The user is the key, eliminating the need to carry any keys or mobile phone and requiring no active operation. This completely solves pain points such as "forgetting to bring keys" or "phone running out of battery," and its convenience far exceeds existing solutions.

[0038] 4. The seamless entry function is decoupled from the security and video recording alarm functions, allowing the core module to operate independently, reducing system integration costs and improving modularity.

[0039] 5. Temporary identity files are used to manage unfamiliar individuals, linking identities across different instances, and automatically cleaning them up every 24 hours. A sliding window is used to analyze suspicious behavior, while also preventing storage overload.

[0040] 6. Gait features are self-learned and updated to adapt to natural changes in the body posture of family members, maintaining a high recognition accuracy over the long term.

[0041] 7. The sensing module can be installed without damage to the vehicle body, load-bearing frame and anti-collision structure, and is compatible with pre-installed mass production and aftermarket non-destructive installation.

[0042] 8. The system has dynamic motion adaptation capabilities. The whitelist has gait models at various speeds pre-stored. Combined with speed adaptive extraction, dynamic time warping, speed extrapolation, and multi-level threshold fusion, it can accurately identify family members running fast, while effectively filtering out false triggers from non-whitelisted personnel.

[0043] Industrial applicability This invention belongs to the field of in-vehicle sensorless perception for intelligent connected vehicles. It features high hardware integration, small size, controllable mass production cost, strong installation adaptability, and can be mass-produced industrially. It can be widely used in various fuel passenger vehicles, new energy passenger vehicles as original factory standard equipment, and aftermarket intelligent upgrade and modification scenarios.

Claims

1. A vehicle-mounted non-intrusive entry system based on gait and microphysiological characteristics, characterized in that, include: The non-contact human feature perception module, without any facial recognition cameras, collects the walking gait, body contours, and micro-physiological features of people around the vehicle. Family members do not need to carry any physical keys, mobile phones, NFC cards, or other electronic devices; their own gait and micro-physiological features serve as the sole key to identity authentication. The local AI feature modeling and comparison module generates a unique digital identity code for each person based on the collected gait and micro-physiological features, and establishes and stores a fixed feature whitelist for family members. For ordinary strangers passing by, only a short-term temporary feature cache is provided, with an automatic timeout clearing mechanism. The non-contact entry control module is linked to the vehicle's original door lock electronic control system. When a registered family member is detected approaching a preset recognition distance, feature comparison and verification are automatically completed, and the car door is unlocked non-contactly, allowing family members to enter the vehicle directly without keys or mobile phone operation.

2. The system according to claim 1, characterized in that: The non-contact human feature sensing module employs non-contact sensing technology, including but not limited to one or more of millimeter-wave radar, ultrasonic sensors, UWB radar, and Wi-Fi sensing modules, to extract the walking gait, body contour, and micro-physiological features.

3. The system according to claim 1, characterized in that: The non-intrusive human feature sensing module is hidden inside the plastic exterior or interior cavity of the vehicle, without damaging the vehicle's load-bearing frame, and can penetrate the plastic trim to collect sensing signals.

4. The system according to claim 1, characterized in that: The local AI feature modeling and comparison module generates a unique digital identity ID for each person, which can accurately distinguish individual identities when multiple people approach the vehicle at the same time. All feature comparisons and identity determinations are completed on the vehicle's local edge computing, and the feature encoding data is stored locally and not uploaded to the cloud.

5. The system according to claim 1, characterized in that: The seamless entry control module enables seamless recognition from a distance of 3 to 5 meters, allowing family members to automatically unlock and open the door when they approach the vehicle from any direction.

6. The system according to claim 1, characterized in that: The microphysiological characteristics include at least one or more of respiration, heartbeat, and limb micro-vibrations.

7. The system according to claim 1, characterized in that: It also includes a stranger behavior intelligent identification module, which counts the frequency of strangers appearing and their movement trajectories based on temporary cached personnel characteristics, and judges them as suspicious reconnaissance behavior when preset abnormal conditions are met.

8. The system according to claim 7, characterized in that: It also includes an intelligent video recording and alarm module, which automatically starts recording external video and internal video when suspicious behavior is detected, and triggers light prompts, horn sounds and remote alarm push notifications to mobile phones.

9. The system according to claim 1, characterized in that: The system has a gait feature self-learning and updating function, which can dynamically and adaptively update the feature whitelist based on the natural changes in the gait and body posture of family members, and continuously maintain the recognition accuracy.

10. The system according to claim 1, characterized in that: The system is compatible with various types of fuel-powered passenger vehicles, new energy passenger vehicles, SUVs, and commercial vehicles. It can be installed as an original equipment manufacturer (OEM) module or added as an aftermarket module without any damage.