Intelligent gate based on radar

Abstract

The invention discloses an intelligent gate machine based on radar. The intelligent gate machine comprises a radar monitoring module, a data processing module, a passing logic control module and a fan door module, the invention relates to the technical field of intelligent security and protection. According to the intelligent gate based on the radar, high-precision three-dimensional perception is carried out through the radar, human face or biological information does not need to be collected, non-contact privacy protection is achieved, point cloud data are processed, whether a target is a person or an object can be judged, the behavior intention of the target is judged by predicting the moving track of the target, and the target behavior intention is more accurate. The system can also automatically keep the gate machine open and give an alarm when falling and retention are detected, realizes real-time response of abnormal behaviors, can realize target identification of adults and children, is matched with judgment of the moving speed of the adults, dynamically adjusts a safety area, ensures that a target person smoothly and efficiently passes through the gate machine, and is matched with real-time monitoring of the feedback force of a leaf door. The risk of clamping people is effectively avoided, and intelligent and safe operation of the gate is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent security technology, specifically to a radar-based intelligent gate. Background Technology

[0002] A turnstile is a channel management device used to manage pedestrian flow and regulate pedestrian access. It can be used for entrance and exit management in various situations.

[0003] Traditional turnstiles use infrared photoelectric sensor technology, which has problems such as large blind spots, susceptibility to environmental interference, and inability to identify the movement trajectory of people. Existing camera-based vision solutions pose a risk of privacy leakage, and the recognition rate drops significantly in low light and strong light environments. Furthermore, in rail transit turnstiles, current technology cannot distinguish between adults and children, leading to the risk of people being trapped when they enter the turnstile one after the other. At the same time, the lack of recognition of abnormal behaviors such as people lingering or falling causes certain safety hazards when using turnstiles.

[0004] Based on the above information, a radar-based intelligent gate is proposed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a radar-based intelligent gate, which solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a radar-based intelligent gate, comprising: A radar monitoring module is installed at both ends and inside the gate to collect point cloud data of the target in real time. The data processing module interfaces with the radar monitoring module and is used to distinguish people and objects based on point cloud data, predict the future movement trajectory of the target, and output behavioral intent tags. The passage logic control module is connected to the data processing module and is used to dynamically adjust the safety zone range according to the target's height and speed, and to issue an audible and visual alarm when the target's behavior is abnormal. The fan gate module interfaces with the access logic control module and is used to receive and execute instructions issued by the access logic control module.

[0007] The present invention is further configured such that the method for distinguishing people and objects based on point cloud data includes: Micro-Doppler feature extraction: After pulse compression of point cloud data, time-frequency spectrum is obtained by short-time Fourier transform. The frequency band corresponding to chest cavity micro-movement is extracted from the time-frequency spectrum. Respiration: 0.2Hz~2Hz, heartbeat: 1Hz~2.5Hz. Respiratory frequency and heartbeat frequency are calculated by peak detection algorithm. Point cloud density analysis involves removing ground point clouds, rasterizing the remaining point clouds into voxels (5cm×5cm×5cm), counting the number of points within each voxel, calculating the global point cloud density, and classifying human and object points. The point cloud density for human points is less than 0.1 points / cm², while the point cloud density for object points is greater than 0.5 points / cm².

[0008] The present invention is further configured such that the method for predicting the future movement trajectory of the target includes: The Kalman filter algorithm is used to calculate the target's velocity and acceleration in real time based on the coordinates returned by the radar, in order to predict the trajectory of the target within the next 2 seconds.

[0009] The present invention is further configured such that the behavioral intent label includes normal passage, lingering, falling, and reverse entry.

[0010] The present invention is further configured such that the method of outputting the behavioral intent label includes: Collect radar point cloud data of 10 consecutive frames containing different behavioral intent labels, project them onto a 3D voxel grid, count the number of point clouds in each voxel, generate a voxel density map, and then perform data augmentation processing. The processing model is constructed based on a 3D convolutional neural network. The input layer receives the preprocessed voxel density tensor and encodes the spatial distribution and dynamic changes. After feature processing through three groups of convolutional layers, the behavioral intent label is output through two fully connected layers.

[0011] The present invention is further configured such that the method of dynamically adjusting the safe zone range according to the target height and speed includes: The height of the target person is determined based on point cloud data. If the height is less than 1.2m, it is determined that there is a child among the target people. The inter-frame difference method is used to calculate the target's movement speed in real time, and a walking threshold is set to determine whether the walking speed is fast or slow. Determine the radius of the safe zone for adults at slow speed, and dynamically adjust the size of the safe zone in fast mode. The adjustment formula is as follows:

[0012] In the formula, The adjusted radius of the safety zone. The radius of the safe zone for adults at slow speed. The target movement speed; When a child is present, the radius of the safe zone is dynamically adjusted and increased to 130% of that of an adult.

[0013] The present invention is further configured such that the method for determining the abnormality of the target behavior is: When the absolute value of the target's acceleration is detected to be greater than 3 m / s², it is determined that the target has fallen. When the target remains stationary for more than 5 seconds, it is determined that the target is stuck. At this time, an audible and visual alarm is triggered and the door remains open.

[0014] The present invention is further configured such that: the fan door module is also used to detect the feedback force of the fan door, and when the feedback force is greater than a preset target value, control the fan door to reverse by a preset angle.

[0015] This invention provides a radar-based intelligent gate. It has the following advantages: This invention utilizes radar for high-precision three-dimensional perception, adapting to complex environments and supporting all-weather operation. It eliminates the need to collect facial or biometric information, achieving contactless privacy protection. Furthermore, by processing point cloud data, it can determine whether the target is a person or an object, and by predicting the target's movement trajectory, it can determine the target's behavioral intentions. It can also automatically keep the gate open and issue an alarm when a fall or lingering is detected, enabling real-time response to abnormal behavior. Simultaneously, it can identify both adults and children, dynamically adjusting the safety zone based on the adult's movement speed to ensure the target person can pass through the gate smoothly and efficiently. Combined with real-time monitoring of the gate's feedback force, it effectively avoids the risk of people being trapped, achieving intelligent and safe operation of the gate. Attached Figure Description

[0016] Figure 1 This is a system architecture block diagram of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0018] Please see Figure 1 The present invention provides the following technical solution: a radar-based intelligent gate, comprising a radar monitoring module, a data processing module, a passage logic control module, and a gate module. The radar monitoring module is deployed at both ends and inside the gate to collect point cloud data of targets in real time. Specifically, two millimeter-wave radar arrays are used in the 24GHz and 77GHz frequency bands respectively, with a horizontal coverage angle ≥120°, a vertical coverage angle ≥15°, a resolution ≤5cm, and support for simultaneous tracking of ≥8 targets. The entrance-side radar is installed on the top of the gate column with an inclination angle of 15°, covering the entrance area of ​​the passage. The exit-side radar is installed at the bottom of the gate column with an inclination angle of 30°, covering the exit area of ​​the passage and the gate area. A single-line lidar is deployed at the top center of the channel, with a vertical scanning angle of ±15° and a sampling frequency of 50Hz, to detect the vertical target distribution within the channel.

[0019] As a preferred solution, the data processing module interfaces with the radar monitoring module. The data processing module is used to distinguish between people and objects based on point cloud data, specifically in the following ways: Micro-Doppler feature extraction: After pulse compression of point cloud data, time-frequency spectrum is obtained by short-time Fourier transform. The frequency band corresponding to chest cavity micro-movement is extracted from the time-frequency spectrum. Respiration: 0.2Hz~2Hz, heartbeat: 1Hz~2.5Hz. Respiratory frequency and heartbeat frequency are calculated by peak detection algorithm. Point cloud density analysis involves removing ground point clouds, rasterizing the remaining point clouds into voxels (5cm×5cm×5cm), counting the number of points within each voxel, calculating the global point cloud density, and classifying human and object points. The point cloud density for human points is less than 0.1 points / cm², while the point cloud density for object points is greater than 0.5 points / cm².

[0020] Predicting the target's future trajectory: Using the Kalman filter algorithm, the target's velocity and acceleration are calculated in real time based on the coordinates returned by the radar to predict its trajectory within the next 2 seconds.

[0021] Output behavioral intent labels, specifically in the following ways: Ten consecutive frames of radar point cloud data containing different behavioral intent labels were collected and projected onto a three-dimensional voxel grid. The number of point clouds in each voxel was counted to generate a voxel density map. The data was then augmented. The behavioral intent labels included normal passage, lingering, falling, and intrusion in the opposite direction. The processing model is constructed based on a 3D convolutional neural network. The input layer receives the preprocessed voxel density tensor and encodes the spatial distribution and dynamic changes. After feature processing through three groups of convolutional layers, the behavioral intent label is output through two fully connected layers.

[0022] As a detailed explanation, the input layer receives the preprocessed voxel density tensor (16×16×16×10) and simultaneously encodes the spatial distribution and dynamic changes. Convolutional layer group: C1 layer: Uses 32 3×3×3 convolutional kernels to scan the input data with a stride of 1, extracting primary spatiotemporal features (such as local density changes in point clouds), and outputting a feature map with a size of 16×16×16×32; C2 layer: Uses 64 3×3×3 convolutional kernels with a stride of 2 for downsampling, focusing on mesoscale features (such as limb movement patterns), and outputs a feature map size of 8×8×8×64; C3 layer: Uses 128 3×3×3 convolutional kernels with a stride of 2 for further downsampling to capture high-order semantic features (such as overall behavioral patterns), and outputs a feature map size of 4×4×4×128; Fully connected layer: FC1 layer: Flattens the output of the convolutional layer into a 512-dimensional vector, and prevents overfitting by using a Dropout rate of 0.5; FC2 layer: Outputs the probability distribution of 4 types of traffic intention labels.

[0023] As a preferred solution, the access logic control module interfaces with the data processing module. The access logic control module is used to dynamically adjust the safe zone range based on the target's altitude and speed, specifically including: The height of the target person is determined based on point cloud data. If the height is less than 1.2m, it is determined that there is a child among the target people. The inter-frame difference method is used to calculate the target's movement speed in real time, and a walking threshold is set to determine whether the walking speed is fast or slow. Determine the radius of the safe zone for adults at slow speed, and dynamically adjust the size of the safe zone in fast mode. The adjustment formula is as follows:

[0024] In the formula, The adjusted radius of the safety zone. The radius of the safe zone for adults at slow speed. The target movement speed; When a child is present, the radius of the safe zone is dynamically adjusted and increased to 130% of that of an adult.

[0025] As a detailed explanation, in order to achieve functional identification of the radar monitoring area, the radar detection area is divided into a two-dimensional grid. The point cloud is projected onto the ground coordinate system and divided into grids of fixed size. Multiple grids of fixed size are merged into a large area, namely the safe zone.

[0026] When the target's behavior is abnormal, an audible and visual alarm will be triggered. Specifically, if the absolute value of the target's acceleration is detected to be greater than 3 m / s², it is determined that the target has fallen. If the target remains stationary for more than 5 seconds, it is determined that the target has been lingering. At this time, the audible and visual alarm will be triggered and the door will remain open.

[0027] As a preferred option, the gate module interfaces with the access logic control module to receive and execute instructions issued by the access logic control module.

[0028] To further explain, the fan door module is also used to detect the feedback force of the fan door, and when the feedback force is greater than the preset target value, it controls the fan door to reverse at a preset angle to avoid the situation of pinching and injuring people.

[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A radar-based intelligent gate, characterized in that, include: A radar monitoring module is installed at both ends and inside the gate to collect point cloud data of the target in real time. The data processing module interfaces with the radar monitoring module and is used to distinguish people and objects based on point cloud data, predict the future movement trajectory of the target, and output behavioral intent tags. The passage logic control module is connected to the data processing module and is used to dynamically adjust the safety zone range according to the target's height and speed, and to issue an audible and visual alarm when the target's behavior is abnormal. The fan gate module interfaces with the access logic control module and is used to receive and execute instructions issued by the access logic control module.

2. The radar-based intelligent gate according to claim 1, characterized in that, The method for distinguishing people and objects based on point cloud data includes: Micro-Doppler feature extraction: After pulse compression of point cloud data, time-frequency spectrum is obtained by short-time Fourier transform. The frequency band corresponding to chest cavity micro-movement is extracted from the time-frequency spectrum. Respiration: 0.2Hz~2Hz, heartbeat: 1Hz~2.5Hz. Respiratory frequency and heartbeat frequency are calculated by peak detection algorithm. Point cloud density analysis involves removing ground point clouds, rasterizing the remaining point clouds into voxels (5cm×5cm×5cm), counting the number of points within each voxel, calculating the global point cloud density, and classifying human and object points. The point cloud density for human points is less than 0.1 points / cm², while the point cloud density for object points is greater than 0.5 points / cm².

3. The radar-based intelligent gate according to claim 1, characterized in that, The methods for predicting the future movement trajectory of the target include: The Kalman filter algorithm is used to calculate the target's velocity and acceleration in real time based on the coordinates returned by the radar, in order to predict the trajectory of the target within the next 2 seconds.

4. The radar-based intelligent gate according to claim 1, characterized in that, The behavioral intent labels include normal passage, lingering, falling, and intrusion in the opposite direction.

5. A radar-based intelligent gate according to claim 4, characterized in that, The methods for outputting behavioral intent labels include: Collect radar point cloud data of 10 consecutive frames containing different behavioral intent labels, project them onto a 3D voxel grid, count the number of point clouds in each voxel, generate a voxel density map, and then perform data augmentation processing. The processing model is constructed based on a 3D convolutional neural network. The input layer receives the preprocessed voxel density tensor and encodes the spatial distribution and dynamic changes. After feature processing through three groups of convolutional layers, the behavioral intent label is output through two fully connected layers.

6. A radar-based intelligent gate according to claim 1, characterized in that, The method of dynamically adjusting the safe zone range based on target altitude and speed includes: The height of the target person is determined based on point cloud data. If the height is less than 1.2m, it is determined that there is a child among the target people. The inter-frame difference method is used to calculate the target's movement speed in real time, and a walking threshold is set to determine whether the walking speed is fast or slow. Determine the radius of the safe zone for adults at slow speed, and dynamically adjust the size of the safe zone in fast mode. The adjustment formula is as follows: ; In the formula, The adjusted radius of the safety zone. The radius of the safe zone for adults at slow speed. The target movement speed; When a child is present, the radius of the safe zone is dynamically adjusted and increased to 130% of that of an adult.

7. A radar-based intelligent gate according to claim 1, characterized in that, The method for determining abnormal target behavior is as follows: When the absolute value of the target's acceleration is detected to be greater than 3 m / s², it is determined that the target has fallen. When the target remains stationary for more than 5 seconds, it is determined that the target is stuck. At this time, an audible and visual alarm is triggered and the door remains open.

8. A radar-based intelligent gate according to claim 1, characterized in that, The fan door module is also used to detect the feedback force of the fan door, and when the feedback force is greater than the preset target value, control the fan door to reverse by a preset angle.

Images