Millimeter wave radar-based close-range newly generated vehicle target positioning and tracking method and system

Abstract

The application discloses a kind of near distance newly generated vehicle target positioning tracking method and system based on millimeter wave radar, it is related to vehicle target tracking technical field, to solve the problem that the initial state estimation of new target is not accurate in the existing target tracking method, leading to tracking failure.The technical points of the present application include: clustering the preprocessed point cloud data, obtaining dynamic target point cloud cluster;Using right angle corner detection method or long side directionality detection method to extract features from dynamic target point cloud cluster to obtain the initial state of dynamic target;The initial state of dynamic target is input into the target tracking algorithm to continuously track the dynamic target and predict its motion state.The initial state of the new target obtained by the present application has higher precision, and the subsequent tracking filter can converge faster and stabilize, thereby improving the spatial position accuracy, speed accuracy and tracking stability of target tracking.

Description

Technical Field

[0001] This invention relates to the field of vehicle target tracking technology, specifically to a method and system for locating and tracking newly generated vehicle targets at close range based on millimeter-wave radar. Background Technology

[0002] An intelligent vehicle system consists of four parts: environmental perception, localization, path planning, and control. It is a comprehensive intelligent system. The environmental perception system acquires information about the surrounding environment through sensors mounted on the intelligent vehicle, providing a reliable basis for the vehicle's subsequent localization and path planning. Localization is achieved by using a navigation system installed on the intelligent vehicle to obtain its own position, heading, and other information. If all four parts meet the requirements of an intelligent vehicle, autonomous driving can be realized.

[0003] Currently, intelligent vehicles are allowed to test their safety in most real-world road environments, but absolute safety cannot be guaranteed, and traffic accidents are still possible. Therefore, environmental perception systems are crucial, and the quality of perception algorithms directly impacts the safety of intelligent vehicles. To achieve comprehensive and redundant perception of the surrounding environment, most intelligent vehicles are equipped with sensors such as LiDAR, cameras, millimeter-wave radar, and ultrasonic radar. Obstacle detection and tracking are the most important components of the perception system, and currently, there are two methods for detection and tracking: one is vision-based; the other is based on radar point cloud data. Vision-based target detection and tracking are susceptible to adverse weather conditions such as rain, snow, and fog, and are also sensitive to lighting conditions. In target detection and tracking based on radar point cloud data, the initial state of a new dynamic target that has just entered the radar's field of view is unknown. Based solely on the basic information of the observed point cloud cluster samples, it is impossible to accurately predict the target's initial state in a very short time. This leads to inaccurate determination of the initial velocity, initial spatial position, and heading angle of the new dynamic target, making it difficult to determine the initial value of the tracking filter in subsequent target tracking. This can easily cause the tracking filter to require a long convergence time, or even directly cause the filter to diverge and fail to converge. Consequently, the new target cannot be associated with the correct point cloud measurement sample, ultimately leading to a decrease in the accuracy and confidence of target tracking, or even the loss of the tracked target. Summary of the Invention

[0004] To address this, the present invention proposes a method and system for short-range newly generated vehicle target localization and tracking based on millimeter-wave radar, in an attempt to solve or at least alleviate at least one of the problems mentioned above.

[0005] According to one aspect of the present invention, a method for near-range newly generated vehicle target localization and tracking based on millimeter-wave radar is provided, the method comprising the following steps:

[0006] Step 1: Acquire point cloud data collected by millimeter-wave radar;

[0007] Step 2: Preprocess the point cloud data;

[0008] Step 3: Cluster the preprocessed point cloud data to obtain dynamic target point cloud clusters;

[0009] Step 4: Use the right-angle corner detection method or the long-side directionality detection method to extract features from the dynamic target point cloud cluster in order to obtain the initial state of the dynamic target;

[0010] Step 5: Input the initial state of the dynamic target into the target tracking algorithm to continuously track the dynamic target and predict its motion state.

[0011] Furthermore, the initial state of the dynamic target includes initial position, initial velocity, and initial heading angle.

[0012] Furthermore, the preprocessing in step two includes coordinate transformation, which transforms the initial point cloud coordinates to a coordinate system with the center of the vehicle's rear axle as the origin.

[0013] Furthermore, the specific process of extracting features from the dynamic target point cloud cluster using the right-angle corner detection method in step four to obtain the initial state of the dynamic target includes:

[0014] In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest corresponding azimuth angles within the field of view of the millimeter-wave radar are respectively used as the starting point A and the ending point B for right-angle point detection and search.

[0015] For any point N in the point cloud cluster other than the starting point A and the ending point B, i , will N i Connect the starting point A and the ending point B to obtain line segment N. i A and N i B;

[0016] With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; Traverse both sets of point clouds and calculate all distances d iA and all distances d iB Summing gives D isum ;

[0017] N-2 Disum Sort the data and select the corresponding D. isum The smallest n points N j Where 3≤n≤N-2, j=1,2…n; calculate line segment AN j With N j The angle between points B is selected, and the point closest to a right angle among the n angles is chosen as the characteristic right-angle point of the dynamic target point cloud cluster.

[0018] The extracted right-angled corner points are used as the same-side corner points of the initial position target box of the dynamic target. The right-angled corner points are connected to the starting point A to form line segment A′, and the right-angled corner points are connected to the ending point B to form line segment B′. The longer side of line segment A′ and line segment B′ is selected as the long side of the initial position target box of the dynamic target. The direction of the long side of the target box is the initial heading angle of the dynamic target. The initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

[0019] Furthermore, the specific process of extracting features from the dynamic target point cloud cluster using the long-side directionality detection method in step three to obtain the initial state of the dynamic target includes:

[0020] The long-side directionality detection method uses a standard deviation ellipse to measure the directionality of the spatial distribution of a point cloud cluster. In a dynamic target point cloud cluster containing N points, the standard deviation ellipse of the point cloud cluster is calculated. The origin coordinates (x0, y0) of the standard deviation ellipse are represented as follows:

[0021]

[0022] In the formula, x i y i Represents point cloud coordinates; This represents the average coordinates of N points; N represents the total number of points in the point cloud.

[0023] The major axis direction angle θ of the standard deviation ellipse is expressed as:

[0024]

[0025] In the formula, This represents the coordinates of the origin (x0, y0) and the coordinates of the point cloud (x0, y0). i y i The difference;

[0026] The length σ of the major axis of the standard deviation ellipse x and minor axis length σ y Represented as:

[0027]

[0028]

[0029] The point closest to the origin of the polar coordinate system with the rear axle center as the origin is selected from the starting and ending points of the major axis segment of the standard deviation ellipse as the corner point of the initial position target box of the dynamic target; the direction angle θ of the major axis of the standard deviation ellipse is used as the initial heading angle of the dynamic target, and the initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

[0030] Furthermore, after extracting features from the dynamic target point cloud cluster using the right-angle corner detection method or the long-side directionality detection method, the extracted point cloud features are evaluated to determine the final initial state of the dynamic target:

[0031] In the right-angle corner detection method, when the longer side of the target bounding box is greater than a preset value and the right-angle fitting variance D of the feature right-angle corner points is... var If the value is less than a preset value, the initial state of the dynamic target obtained after feature extraction using the right-angle corner detection method is deemed usable; the right-angle fitting variance D var The calculation formula is:

[0032]

[0033] In the long-side directionality detection method, when the length of the major axis of the standard deviation ellipse is greater than a preset value and the ratio of the major axis to the minor axis of the standard deviation ellipse is greater than a preset value, it is determined that the initial state of the dynamic target obtained after feature extraction using the long-side directionality detection method is usable.

[0034] Furthermore, the target tracking algorithm employs Kalman filtering to perform target motion estimation and updating.

[0035] According to another aspect of the present invention, a near-range newly generated vehicle target localization and tracking system based on millimeter-wave radar is provided, the system comprising:

[0036] The data acquisition module is configured to acquire point cloud data collected by millimeter-wave radar;

[0037] A preprocessing module is configured to preprocess the point cloud data, the preprocessing including coordinate transformation, the coordinate transformation being to transform the initial point cloud coordinates to a coordinate system with the center of the rear axle of the vehicle as the origin;

[0038] The point cloud clustering module is configured to cluster preprocessed point cloud data to obtain dynamic target point cloud clusters.

[0039] The feature extraction module is configured to extract features from the dynamic target point cloud cluster using a right-angle corner detection method or a long-side directionality detection method to obtain the initial state of the dynamic target; the initial state of the dynamic target includes the initial position, initial velocity, and initial heading angle;

[0040] The target tracking module is configured to input the initial state of a dynamic target into the target tracking algorithm to continuously track the dynamic target and predict its motion state.

[0041] Furthermore, the specific process of using the right-angle corner detection method to extract features from the dynamic target point cloud cluster in the feature extraction module to obtain the initial state of the dynamic target includes:

[0042] In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest corresponding azimuth angles within the field of view of the millimeter-wave radar are respectively used as the starting point A and the ending point B for right-angle point detection and search.

[0043] For any point N in the point cloud cluster other than the starting point A and the ending point B, i , will N i Connect the starting point A and the ending point B to obtain line segment N. i A and N i B;

[0044] With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; Traverse both sets of point clouds and calculate all distances d iA and all distances d iB Summing gives D isum ;

[0045] N-2 D isum Sort the data and select the corresponding D. isum The smallest n points N j Where 3≤n≤N-2, j=1,2…n; calculate line segment AN j With N j The angle between points B is selected, and the point closest to a right angle among the n angles is chosen as the characteristic right-angle point of the dynamic target point cloud cluster.

[0046] The extracted right-angled corner points are used as the same-side corner points of the initial position target box of the dynamic target. The right-angled corner points are connected to the starting point A to form line segment A′, and the right-angled corner points are connected to the ending point B to form line segment B′. The longer side of line segment A′ and line segment B′ is selected as the long side of the initial position target box of the dynamic target. The direction of the long side of the target box is the initial heading angle of the dynamic target. The initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

[0047] Furthermore, the specific process of using the long-side directionality detection method to extract features from the dynamic target point cloud cluster in the feature extraction module to obtain the initial state of the dynamic target includes:

[0048] The long-side directionality detection method uses a standard deviation ellipse to measure the directionality of the spatial distribution of a point cloud cluster. In a dynamic target point cloud cluster containing N points, the standard deviation ellipse of the point cloud cluster is calculated. The origin coordinates (x0, y0) of the standard deviation ellipse are represented as follows:

[0049]

[0050] In the formula, x i y i Represents point cloud coordinates; This represents the average coordinates of N points; N represents the total number of points in the point cloud.

[0051] The major axis direction angle θ of the standard deviation ellipse is expressed as:

[0052]

[0053] In the formula, This represents the coordinates of the origin (x0, y0) and the coordinates of the point cloud (x0, y0). i y i The difference;

[0054] The length σ of the major axis of the standard deviation ellipse x and minor axis length σ y Represented as:

[0055]

[0056]

[0057] The point closest to the origin of the polar coordinate system with the rear axle center as the origin is selected from the starting and ending points of the major axis segment of the standard deviation ellipse as the corner point of the initial position target box of the dynamic target; the direction angle θ of the major axis of the standard deviation ellipse is used as the initial heading angle of the dynamic target, and the initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

[0058] The beneficial technical effects of this invention are:

[0059] This invention proposes a method and system for near-range newly generated vehicle target localization and tracking based on millimeter-wave radar. It acquires point cloud data collected by millimeter-wave radar, preprocesses the point cloud data, and clusters the preprocessed point cloud data to obtain dynamic target point cloud clusters. Features are extracted from the dynamic target point cloud clusters using right-angle corner detection or long-side directionality detection methods. The usability of these features is further evaluated, and the extracted features are applied to the rapid localization of new targets, thereby obtaining the initial state of the dynamic target. Compared to traditional target tracking methods, the initial state of the newly generated target obtained by this invention has higher accuracy, and subsequent tracking filters can converge and stabilize faster, thus improving the spatial position accuracy, velocity accuracy, and tracking stability of the target. Attached Figure Description

[0060] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0061] Figure 1 This is a flowchart of a short-range newly generated vehicle target localization and tracking method based on millimeter-wave radar, as described in an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of a point cloud sample acquired by radar in an embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram illustrating the spatial location distribution characteristics of vehicle target point cloud clusters in an embodiment of the present invention.

[0064] Figure 4 This is a schematic diagram of the right-angle corner detection method in an embodiment of the present invention.

[0065] Figure 5 This is a schematic diagram of the long side directionality detection method in an embodiment of the present invention.

[0066] Figure 6 This is a schematic diagram of the target tracking process in an embodiment of the present invention.

[0067] Figure 7 This is a schematic diagram of the structure of the near-range newly generated vehicle target localization and tracking system based on millimeter-wave radar according to an embodiment of the present invention. Detailed Implementation

[0068] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are given merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make this disclosure more thorough and complete, and to fully convey the scope of this disclosure to those skilled in the art.

[0069] Those skilled in the art will recognize that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software. It should be understood herein that any number of elements in the accompanying drawings is for illustrative purposes only and not as a limitation, and any naming is for distinction only and has no limiting meaning.

[0070] The limited field of view for close-range target detection and the small Doppler component of targets moving in the normal direction make it difficult to determine the motion trend. Due to the limitation of the field of view (FOV), the sensor can often only observe a limited local area for close-range rigid targets. Furthermore, when the target's motion direction is nearly perpendicular to the sensor's normal direction, the Doppler component generated by the target's motion is very small. This makes it impossible to make a relatively accurate estimate of the initial motion state of a new target under conventional tracking procedures. Insufficient estimation of the initial state will affect the convergence speed and tracking accuracy of subsequent tracking.

[0071] False alarm points or wheel points are prone to appear in the radar's observation point cloud of the near-range target outline. The velocity dimension quality of the observation points degrades, directly affecting the tracking quality. The point cloud detected by the sensor is mainly divided into vehicle outline points and wheel points. In the extended target tracking process of near-range targets, when the wheel part enters the FOV, the proportion of wheel points in the vehicle's detection points is relatively high, and some wheel points will be within the clustering range of vehicle outline points. These wheel points will have a certain impact on the vehicle tracking model. For example, when tracking a target that enters the FOV at a constant speed, due to the large number of wheel points mixed in with the observation values, the velocity dimension changes significantly in a short time. The motion model will estimate that the target moving at a constant speed in a straight line has entered a high maneuver state at this moment, so the heading angle of the target tracking will be shifted, affecting the overall tracking accuracy of the target.

[0072] Therefore, this invention proposes a method and system for locating and tracking newly generated vehicle targets at close range based on millimeter-wave radar. By extracting features from the point cloud clusters associated with the newly generated target and evaluating the usability of the features, the extracted features are then applied to the rapid location of the new target. This enables the accurate and rapid determination of the initial state of the newly generated vehicle target at close range, thereby improving the spatial position accuracy, velocity accuracy, and tracking stability of the target tracking.

[0073] This invention proposes a method for near-range newly generated vehicle target localization and tracking based on millimeter-wave radar, such as... Figure 1 As shown, the method includes the following steps:

[0074] Step 1: Acquire point cloud data collected by millimeter-wave radar;

[0075] Step 2: Preprocess the point cloud data;

[0076] Step 3: Cluster the preprocessed point cloud data to obtain dynamic target point cloud clusters;

[0077] Step 4: Use the right-angle corner detection method or the long-side directionality detection method to extract features from the dynamic target point cloud cluster to obtain the initial state of the dynamic target; the initial state of the dynamic target includes the initial position, initial velocity, and initial heading angle;

[0078] Step 5: Input the initial state of the dynamic target into the target tracking algorithm to continuously track the dynamic target and predict its motion state.

[0079] In step one, point cloud data collected by millimeter-wave radar is acquired, for example as follows: Figure 2 As shown in step two, after acquiring the point cloud data, it is preprocessed. The preprocessing includes coordinate transformation and filtering. The coordinate transformation is to uniformly transform the initial point cloud into a coordinate system with the center of the vehicle's rear axle as the origin. This is because the initial point cloud coordinates are in a coordinate system with the millimeter-wave radar installation position as the origin, which is different from the coordinate system actually required for vehicle radar target tracking. Therefore, it is necessary to uniformly transform the initial point cloud into a coordinate system including polar coordinates and Cartesian coordinates. Due to the measurement principle characteristics of millimeter-wave radar, the initial point cloud contains some noise points (such as noise points). It is necessary to filter out identifiable noise points in the point cloud during preprocessing to reduce false targets generated during subsequent target tracking.

[0080] In step three, the point cloud data is clustered to obtain dynamic target point cloud clusters. The purpose of radar point cloud clustering is to group point clouds with similar spatial locations and motion states into a set of samples, i.e., point cloud clusters. Point clouds within the same cluster have similar properties, while point cloud samples in different clusters are different. As an example, the clustering algorithm can be the K-means algorithm or the DBSCAN algorithm. The DBSCAN algorithm mainly separates high-density regions from the point cloud data, so that points in the same class are very close together, while points in different classes are far apart.

[0081] In step four, the right-angle corner detection method or the long-side directionality detection method are used to extract features from the dynamic target point cloud cluster in order to obtain the initial state of the dynamic target.

[0082] According to embodiments of the present invention, the spatial distribution characteristics of radar point cloud samples on the surface of a near-range vehicle target are analyzed using a right-angle corner detection method or a long-side directionality detection method, and then prediction and fitting are performed to quickly locate the spatial position and heading angle information of the vehicle target. The relative velocity of the target is calculated based on the Doppler velocity dpl and azimuth angle of the point cloud samples. This allows for a more accurate initial state of the emerging target, providing a more accurate starting value for the Kalman filter used for target tracking. This enables the target tracking filter to converge faster in subsequent target tracking, improving the stability and accuracy of target tracking.

[0083] like Figure 3-5 As shown, the point cloud clusters generated by radar waves reflected from close-range vehicle targets vary depending on the vehicle target's position within the millimeter-wave radar's field of view. The surface exposed to the radar beam and reflecting radar waves (hereinafter referred to as the exposure surface) will differ, resulting in different spatial distribution characteristics of the generated radar point clouds. Because the primary reflection exposure surface of close-range vehicle targets differs within the radar's field of view (see...), the spatial distribution characteristics of the generated radar point clouds will also differ. Figure 3 As shown in the figure, the spatial distribution characteristics of the generated point cloud clusters are also different, mainly including L-shaped distribution on the side and bottom and strip distribution with the long side as the main feature. Therefore, the feature extraction methods used in this embodiment of the invention are mainly two: right-angle corner detection method and long side directionality detection method.

[0084] 1) Right-angle corner detection method:

[0085] In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest azimuth angles within the field of view of the millimeter-wave radar are respectively used as the starting point A and the ending point B for right-angle point detection and search.

[0086] For any point N in the point cloud cluster other than the starting point A and the ending point B, i , will N i Connect the starting point A and the ending point B to obtain line segment N. i A and N i B;

[0087] With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; Traverse both sets of point clouds and calculate all distances d iA and all distances d iB Summing gives Disum ;

[0088] N-2 D isum Sort the data and select the corresponding D. isum The smallest n points N j Where 3≤n≤N-2, j=1,2…n; calculate line segment AN j With N j The angle between points B and n points is selected, with the point closest to a right angle being the characteristic right-angle point of the dynamic target point cloud cluster.

[0089] 2) Long-side directionality detection method:

[0090] The standard deviation ellipse is used to measure the directionality of the spatial distribution of point cloud clusters. The standard deviation ellipse of the point cloud cluster is calculated according to the existing formula. To determine the standard deviation ellipse, it is necessary to determine the origin position, the orientation angle of the major axis, and the length of the major and minor axes.

[0091] a) Determine the origin of the standard deviation ellipse. The origin (x0, y0) of the ellipse is determined by the center of the arithmetic mean of the spatial positions of the points in the point cloud cluster:

[0092]

[0093] In the formula, x i y i Represents point cloud coordinates; This represents the average coordinates of N points; N represents the total number of points in the point cloud.

[0094] b) Determine the direction of the standard deviation ellipse, with the major axis as the reference, and true north as 0 degrees.

[0095]

[0096] In the formula, This represents the coordinates of the origin (x0, y0) and the coordinates of the point cloud (x, y0). i、 y i The difference;

[0097] c) Determine the length σ of the major axis of the standard deviation ellipse. x and minor axis length σ y :

[0098]

[0099]

[0100] In the formula, θ represents the direction angle of the major axis of the standard deviation ellipse.

[0101] In this embodiment, preferably, the extracted features are further evaluated for usability. The main purpose is to determine whether the distribution characteristics of the obtained point cloud clusters can reflect the morphological characteristics of the actual vehicle target. Depending on the feature extraction method, this is divided into two aspects: usability evaluation of the right-angled corner points of the target and usability evaluation of the directionality of the target's long side.

[0102] The evaluation criteria for the usability of right-angled corner points mainly include: the number of point cloud clusters, the length of the two right-angled sides fitted, and the fitting variance of the right-angled corner points. Among these, a certain number of point cloud samples are needed in the point cloud cluster to have fitting value. If the number of point cloud samples is too small, the reference value of the fitting result is limited. Depending on the radar resolution, the minimum number of point cloud samples should be selected according to the actual situation, generally between 10 and 40. Since the minimum rigid body volume of ordinary vehicle targets is basically determined, the fitting result is only usable when the longer right-angled side of the fitted right-angled L-shape is greater than a preset value. In this embodiment, the preset value for the longer right-angled side is set to 2.5–3.5 m. The right-angle fitting variance D of the feature right-angled corner points obtained during the right-angled corner point feature extraction process is... var D reflects the degree of fit of the fitting results. var The smaller the value, the better the fit. var The larger the value, the worse the fit, where the variance of the rectangular fit, D, is... var The calculation formula is:

[0103]

[0104] The main usability evaluation criteria for the target's long-side directionality include: the length of the major axis of the standard deviation ellipse and the ratio of the major and minor axes of the standard deviation ellipse. The actual length of the major axis of the standard deviation ellipse reflects the spatial size of the ellipse. When the length of the major axis of the fitted target's long-side standard deviation ellipse is greater than a preset value, it indicates that the spatial distribution of the point cloud cluster generated by the target in the radar field of view is sufficiently large, and can be used as a basis for target directionality analysis. In this embodiment, the preset value can be 2.5-3.5m. The ratio of the major and minor axes of the standard deviation ellipse reflects the directional characteristics of the point cloud cluster. A larger ratio indicates a flatter and more slender ellipse, and the direction angle of the major axis better reflects the spatial distribution of the point cloud. In this embodiment, the preset threshold for the ratio can be set to 3-5. A ratio greater than this preset threshold indicates that the ellipse's directionality is relatively obvious, and the direction angle θ of the major axis can be used to predict the target's heading angle information.

[0105] After feature extraction from the point cloud clusters associated with newly generated vehicle targets at close range, the point cloud features that meet the usability assessment are used for localization of the initial state of the new target: if right-angle corner features are available, the right-angle corner position is used as the corner point on the same side of the new target bounding box (see...). Figure 4To locate the spatial position of the new target, connect the feature right-angled corner point with the starting point A to form line segment A', and connect the feature right-angled corner point with the ending point B to form line segment B'. Select the longer side of line segment A' and line segment B' as the long side of the initial position target box of the dynamic target. Use the direction of the long right-angled side to locate the heading angle of the new target, and use the heading and known point cloud information (Doppler velocity) to deduce the initial velocity vel of the target.

[0106] The target bounding box size of the new target is set with default length, width, and height based on the new target's classification category. The length and width of the bounding box can be adjusted according to the length of the fitted right-angled sides. If the directional feature of the long side is available, the spatial position of the new target is located by taking the point closest to the origin of the polar coordinate system (with the rear axle center as the origin) between the starting and ending points of the major axis segment of the standard deviation ellipse and the point of origin of the standard deviation ellipse (see [link]). Figure 5 The heading angle (heading) of the new target is determined using the direction angle θ of the major axis of the standard deviation ellipse. Similarly, the initial velocity (vel) of the target is deduced from the heading and the known point cloud information (Doppler velocity). The length and width of the target bounding box are then corrected with reference to the length of the major axis.

[0107] If both right-angle corner features and long-side directional features of a point cloud cluster are available, the right-angle corner features are preferred for locating new targets because they provide richer information.

[0108] In step five, the initial state of the dynamic target is input into the target tracking algorithm to continuously track the dynamic target and predict its motion state.

[0109] As an example, the target tracking algorithm uses Kalman filtering to estimate and update the target's motion. After locating the initial motion state and spatial position of a new target based on features, the initial state is written into the target state initial value of the Kalman filter, and the new target is added to the target tracking data stream. Target tracking consists of two steps: data association and state estimation. Data association refers to linking the detection data of the same target in different frames to determine whether the target in the preceding and following frames is the same. State estimation refers to applying a filter to predict the target's motion state and obtain information such as its position, velocity, and heading angle. The specific process is as follows: Figure 6 As shown.

[0110] This invention fully utilizes the abundant point cloud cluster information from near-range radar. By analyzing the possible distribution patterns of point cloud clusters, it fits the L-shaped right-angled sides of the target tracking box exposure surface and the distribution directionality of point cloud clusters in the current time slice, and calculates the degree of matching between the actual point cloud distribution and the fitting result. This allows for prior estimation of the target box boundary and volume, the initial heading angle of the target, etc., and the calculation of the confidence level of the prior estimates. Subsequently, prior information with high confidence can be used in a weighted manner as needed. More comprehensive use of target prior information allows newly generated tracking targets to have higher initial confidence, enabling faster output of high-confidence mature targets. When the confidence level of the prior information is high enough, new targets can directly start as mature targets within a single frame. Furthermore, during the tracking of mature targets, this prior information can also serve as a reference for verification and outlier correction.

[0111] Another embodiment of the present invention proposes a near-range newly generated vehicle target localization and tracking system based on millimeter-wave radar, such as... Figure 7 As shown, the system includes:

[0112] The data acquisition module 710 is configured to acquire point cloud data collected by millimeter-wave radar;

[0113] The preprocessing module 720 is configured to preprocess the point cloud data, the preprocessing including coordinate transformation, the coordinate transformation being to transform the initial point cloud coordinates to a coordinate system with the center of the rear axle of the vehicle as the origin;

[0114] The point cloud clustering module 730 is configured to cluster preprocessed point cloud data to obtain dynamic target point cloud clusters.

[0115] The feature extraction module 740 is configured to extract features from the dynamic target point cloud cluster using a right-angle corner detection method or a long-side directionality detection method to obtain the initial state of the dynamic target; the initial state of the dynamic target includes the initial position, initial velocity, and initial heading angle;

[0116] The target tracking module 750 is configured to input the initial state of a dynamic target into the target tracking algorithm to continuously track the dynamic target and predict its motion state.

[0117] In this embodiment, preferably, the specific process of using the right-angle corner detection method to extract features from the dynamic target point cloud cluster in the feature extraction module 740 to obtain the initial state of the dynamic target includes:

[0118] In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest corresponding azimuth angles within the field of view of the millimeter-wave radar are respectively used as the starting point A and the ending point B for right-angle point detection and search.

[0119] For any point N in the point cloud cluster other than the starting point A and the ending point B, i , will N i Connect the starting point A and the ending point B to obtain line segment N. i A and N i B;

[0120] With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; Traverse both sets of point clouds and calculate all distances d iA and all distances d iB Summing gives D isum ;

[0121] N-2 D isum Sort the data and select the corresponding D. isum The smallest n points N j Where 3≤n≤N-2, j=1,2…n; calculate line segment AN j With N j The angle between points B is selected, and the point closest to a right angle among the n angles is chosen as the characteristic right-angle point of the dynamic target point cloud cluster.

[0122] The extracted right-angled corner points are used as the same-side corner points of the initial position target box of the dynamic target. The right-angled corner points are connected to the starting point A to form line segment A′, and the right-angled corner points are connected to the ending point B to form line segment B′. The longer side of line segment A′ and line segment B′ is selected as the long side of the initial position target box of the dynamic target. The direction of the long side of the target box is the initial heading angle of the dynamic target. The initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

[0123] In this embodiment, preferably, the specific process of using the long-side directionality detection method to extract features from the dynamic target point cloud cluster in the feature extraction module 740 to obtain the initial state of the dynamic target includes:

[0124] The long-side directionality detection method uses a standard deviation ellipse to measure the directionality of the spatial distribution of a point cloud cluster. In a dynamic target point cloud cluster containing N points, the standard deviation ellipse of the point cloud cluster is calculated. The origin coordinates (x0, y0) of the standard deviation ellipse are represented as follows:

[0125]

[0126] In the formula, x i y i Represents point cloud coordinates; This represents the average coordinates of N points; N represents the total number of points in the point cloud.

[0127] The major axis direction angle θ of the standard deviation ellipse is expressed as:

[0128]

[0129] In the formula, This represents the coordinates of the origin (x0, y0) and the coordinates of the point cloud (x0, y0). i y i The difference;

[0130] The length σ of the major axis of the standard deviation ellipse x and minor axis length σ y Represented as:

[0131]

[0132]

[0133] The point closest to the origin of the polar coordinate system with the rear axle center as the origin is selected from the starting and ending points of the major axis segment of the standard deviation ellipse as the corner point of the target frame of the dynamic target. The direction angle θ of the major axis of the standard deviation ellipse is used as the initial heading angle of the dynamic target. The initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

[0134] It should be noted that although several units, modules, or sub-modules are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules described above can be embodied in one module. Conversely, the features and functions of one module described above can be further divided and embodied by multiple modules.

[0135] Furthermore, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0136] While the spirit and principles of the invention have been described with reference to several specific embodiments, it should be understood that the invention is not limited to the disclosed specific embodiments, and the division of aspects does not imply that features in these aspects cannot be combined for benefit; such division is merely for ease of description. The invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A method for short-range newly generated vehicle target localization and tracking based on millimeter-wave radar, characterized in that, Includes the following steps: Step 1: Acquire point cloud data collected by millimeter-wave radar; Step 2: Preprocess the point cloud data; Step 3: Cluster the preprocessed point cloud data to obtain dynamic target point cloud clusters; Step 4: Extract features from the dynamic target point cloud cluster using the right-angle corner detection method or the long-side directionality detection method to obtain the initial state of the dynamic target; after extracting features from the dynamic target point cloud cluster using the right-angle corner detection method or the long-side directionality detection method, evaluate the extracted point cloud features to determine the final initial state of the dynamic target: In the right-angle corner detection method, when the longer side of the target bounding box is greater than a preset value and the right-angle fitting variance D of the feature right-angle corner points is... var If the value is less than a preset value, the initial state of the dynamic target obtained after feature extraction using the right-angle corner detection method is deemed usable; the right-angle fitting variance D var The calculation formula is: ； In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest corresponding azimuth angles within the field of view of the millimeter-wave radar are designated as the starting point A and ending point B for right-angle point detection and search, respectively. All other points are designated as N. i With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; In the long-side directionality detection method, when the length of the major axis of the standard deviation ellipse is greater than a preset value and the ratio of the major axis to the minor axis of the standard deviation ellipse is greater than a preset value, it is determined that the initial state of the dynamic target obtained after feature extraction using the long-side directionality detection method is usable. In this process, after feature extraction from the point cloud cluster, the point cloud features that meet the usability assessment are used to locate the initial state of the new target: if the right-angle corner feature is available, the right-angle corner position is used as the corner point on the same side of the new target bounding box to locate the spatial position of the new target; if the long side directional feature is available, the corner point of the new target bounding box is used as the closest point between the start and end points of the standard deviation ellipse major axis segment and the origin of the polar coordinate system with the rear axle center of the vehicle as the origin to locate the spatial position of the new target; if both the right-angle corner feature and the long side directional feature of the point cloud cluster are available, the right-angle corner feature is used first to locate the new target. Step 5: Input the initial state of the dynamic target into the target tracking algorithm to continuously track the dynamic target and predict its motion state.

2. The method for near-range newly generated vehicle target localization and tracking based on millimeter-wave radar according to claim 1, characterized in that, The initial state of the dynamic target includes its initial position, initial velocity, and initial heading angle.

3. The method for near-range newly generated vehicle target localization and tracking based on millimeter-wave radar according to claim 2, characterized in that, The preprocessing described in step two includes coordinate transformation, which transforms the initial point cloud coordinates to a coordinate system with the center of the vehicle's rear axle as the origin.

4. The method for short-range newly generated vehicle target localization and tracking based on millimeter-wave radar according to claim 3, characterized in that, Step four involves using the right-angle corner detection method to extract features from the dynamic target point cloud cluster to obtain the initial state of the dynamic target. The specific process includes: In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest corresponding azimuth angles within the field of view of the millimeter-wave radar are respectively used as the starting point A and the ending point B for right-angle point detection and search. For any point N in the point cloud cluster other than the starting point A and the ending point B, i , will N i Connect the starting point A and the ending point B to obtain line segment N. i A and N i B; With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; Traverse both sets of point clouds and calculate all distances d iA and all distances d iB Summing gives D isum ; N-2 D isum Sort the data and select the corresponding D. isum The smallest n points N j Where 3≤n≤N-2, j=1,2…n; calculate line segment AN j With N j The angle between points B is selected, and the point closest to a right angle among the n angles is chosen as the characteristic right-angle point of the dynamic target point cloud cluster. The extracted right-angled corner points are used as the same-side corner points of the initial position target box of the dynamic target. The right-angled corner points are connected to the starting point A to form line segment A', and the right-angled corner points are connected to the ending point B to form line segment B'. The longer side of line segment A' and line segment B' is selected as the long side of the initial position target box of the dynamic target. The direction of the long side of the target box is the initial heading angle of the dynamic target. The initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

5. The method for near-range newly generated vehicle target localization and tracking based on millimeter-wave radar according to claim 4, characterized in that, Step four involves using the long-side directionality detection method to extract features from the dynamic target point cloud cluster to obtain the initial state of the dynamic target. The specific process includes: The long-side directionality detection method uses a standard deviation ellipse to measure the directionality of the spatial distribution of a point cloud cluster. In a dynamic target point cloud cluster containing N points, the standard deviation ellipse of the point cloud cluster is calculated. The origin coordinates (x0, y0) of the standard deviation ellipse are represented as follows: ； In the formula, x i、 y i Represents point cloud coordinates; This represents the average coordinates of N points; N represents the total number of points in the point cloud. The major axis direction angle θ of the standard deviation ellipse is expressed as: ； In the formula, This represents the coordinates of the origin (x0, y0) and the coordinates of the point cloud (x, y0). i、 y i The difference; The length σ of the major axis of the standard deviation ellipse x and minor axis length σ y Represented as: ； ； The point closest to the origin of the polar coordinate system with the rear axle center as the origin is selected from the starting and ending points of the major axis segment of the standard deviation ellipse as the corner point of the target frame of the dynamic target. The direction angle θ of the major axis of the standard deviation ellipse is used as the initial heading angle of the dynamic target. The initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

6. The method for near-range newly generated vehicle target localization and tracking based on millimeter-wave radar according to claim 1, characterized in that, The target tracking algorithm uses Kalman filtering to estimate and update the target's motion.

7. A near-range newly generated vehicle target localization and tracking system based on millimeter-wave radar, characterized in that, include: The data acquisition module is configured to acquire point cloud data collected by millimeter-wave radar; A preprocessing module is configured to preprocess the point cloud data, the preprocessing including coordinate transformation, the coordinate transformation being to transform the initial point cloud coordinates to a coordinate system with the center of the rear axle of the vehicle as the origin; The point cloud clustering module is configured to cluster preprocessed point cloud data to obtain dynamic target point cloud clusters. The feature extraction module is configured to extract features from the dynamic target point cloud cluster using either right-angle corner detection or long-side directionality detection to obtain the initial state of the dynamic target. The initial state of the dynamic target includes its initial position, initial velocity, and initial heading angle. After extracting features from the dynamic target point cloud cluster using either the right-angle corner detection or long-side directionality detection method, the extracted point cloud features are evaluated to determine the final initial state of the dynamic target. In the right-angle corner detection method, when the longer side of the target bounding box is greater than a preset value and the right-angle fitting variance D of the feature right-angle corner points is... var If the value is less than a preset value, the initial state of the dynamic target obtained after feature extraction using the right-angle corner detection method is deemed usable; the right-angle fitting variance D var The calculation formula is: ； In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest corresponding azimuth angles within the field of view of the millimeter-wave radar are designated as the starting point A and ending point B for right-angle point detection and search, respectively. All other points are designated as N. i With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; In the long-side directionality detection method, when the length of the major axis of the standard deviation ellipse is greater than a preset value and the ratio of the major axis to the minor axis of the standard deviation ellipse is greater than a preset value, it is determined that the initial state of the dynamic target obtained after feature extraction using the long-side directionality detection method is usable. In this process, after feature extraction from the point cloud cluster, the point cloud features that meet the usability assessment are used to locate the initial state of the new target: if the right-angle corner feature is available, the right-angle corner position is used as the corner point on the same side of the new target bounding box to locate the spatial position of the new target; if the long side directional feature is available, the corner point of the new target bounding box is used as the closest point between the start and end points of the standard deviation ellipse major axis segment and the origin of the polar coordinate system with the rear axle center of the vehicle as the origin to locate the spatial position of the new target; if both the right-angle corner feature and the long side directional feature of the point cloud cluster are available, the right-angle corner feature is used first to locate the new target. The target tracking module is configured to input the initial state of a dynamic target into the target tracking algorithm to continuously track the dynamic target and predict its motion state.

8. The near-range newly generated vehicle target localization and tracking system based on millimeter-wave radar according to claim 7, characterized in that, The feature extraction module uses a right-angle corner detection method to extract features from the dynamic target point cloud cluster to obtain the initial state of the dynamic target. The specific process includes: In a dynamic target point cloud cluster containing N points, the two points with the largest and smallest corresponding azimuth angles within the field of view of the millimeter-wave radar are respectively used as the starting point A and the ending point B for right-angle point detection and search. For any point N in the point cloud cluster other than the starting point A and the ending point B, i , will N i Connect the starting point A and the ending point B to obtain line segment N. i A and N i B; With point N i Using N as the dividing point, the point cloud clusters with values ​​greater than N are... i and less than N i The point cloud is divided into two groups, and calculations are performed on groups greater than N. i In the point cloud group, each point is connected to line segment N. i The distance d from A iA Calculate less than N i In the point cloud group, each point is connected to line segment N. i The distance d from B iB ; Traverse both sets of point clouds and calculate all distances d iA and all distances d iB Summing gives D isum ; N-2 D isum Sort the data and select the corresponding D. isum The smallest n points N j Where 3≤n≤N-2, j=1,2…n; calculate line segment AN j With N j The angle between points B is selected, and the point closest to a right angle among the n angles is chosen as the characteristic right-angle point of the dynamic target point cloud cluster. The extracted right-angled corner points are used as the same-side corner points of the initial position target box of the dynamic target. The right-angled corner points are connected to the starting point A to form line segment A', and the right-angled corner points are connected to the ending point B to form line segment B'. The longer side of line segment A' and line segment B' is selected as the long side of the initial position target box of the dynamic target. The direction of the long side of the target box is the initial heading angle of the dynamic target. The initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

9. The near-range newly generated vehicle target localization and tracking system based on millimeter-wave radar according to claim 8, characterized in that, The feature extraction module uses the long-side directionality detection method to extract features from the dynamic target point cloud cluster to obtain the initial state of the dynamic target. The specific process includes: The long-side directionality detection method uses a standard deviation ellipse to measure the directionality of the spatial distribution of a point cloud cluster. In a dynamic target point cloud cluster containing N points, the standard deviation ellipse of the point cloud cluster is calculated. The origin coordinates (x0, y0) of the standard deviation ellipse are represented as follows: ； In the formula, x i、 y i Represents point cloud coordinates; This represents the average coordinates of N points; N represents the total number of points in the point cloud. The major axis direction angle θ of the standard deviation ellipse is expressed as: ； In the formula, This represents the coordinates of the origin (x0, y0) and the coordinates of the point cloud (x, y0). i、 y i The difference; The length σ of the major axis of the standard deviation ellipse x and minor axis length σ y Represented as: ； ； The point closest to the origin of the polar coordinate system with the rear axle center as the origin is selected from the starting and ending points of the major axis segment of the standard deviation ellipse as the corner point of the initial position target box of the dynamic target; the direction angle θ of the major axis of the standard deviation ellipse is used as the initial heading angle of the dynamic target, and the initial velocity of the dynamic target is obtained by calculating the initial heading angle and the Doppler velocity of the point cloud data.

Images