Road congestion degree evaluation method and device based on laser radar detection

Abstract

The invention discloses a road congestion degree evaluation method and device based on laser radar detection, and relates to the field of vehicle control, and the method comprises the steps: constructing an ENU coordinate system with a preset reference point as a coordinate origin as a reference coordinate system, and calculating an external parameter relation between a main laser radar point cloud coordinate system and the reference coordinate system; drawing and recording a closed boundary of the selected road section on an XOY projection plane of the reference coordinate system; preprocessing the laser radar point cloud, including coordinate transformation and elimination of point cloud outside the boundary of the road section and ground point cloud; obtaining a laser radar vehicle detection method in deep learning; based on a laser radar vehicle detection method in deep learning, all vehicles in a selected road section at a certain moment are detected, the number of the vehicles is counted, and traffic flow data are obtained; the method comprises the following steps: scanning correlation matching vehicles of front and back frames, smoothing vehicle tracks through Kalman filtering, calculating real-time vehicle speeds, and counting an average number and a median of the speeds of all effective vehicles in a road section;

Description

Technical Field

[0001] This application relates to the field of vehicle control, and in particular to a method for assessing road congestion based on lidar detection, a device for assessing road congestion based on lidar detection, electronic equipment, storage media, and vehicle platform. Background Technology

[0002] Accurate assessment of road congestion can help drivers anticipate their routes and assist traffic management departments in understanding traffic conditions. Currently, most methods for automatically assessing road congestion rely on computer vision. However, this method is limited by factors such as lighting conditions and obstructions in outdoor environments, resulting in limitations in its recognition capabilities. LiDAR, compared to cameras, offers advantages such as strong resistance to lighting effects and accurate position measurement, making it more suitable for real-time target detection in open outdoor conditions. By analyzing traffic flow on a specific representative road segment and combining it with vehicle speed information, the degree of road congestion can be assessed. This method boasts advantages such as accurate detection, high real-time performance, and a high degree of automation.

[0003] Compared to the patented method, device, electronic device, and readable storage medium for identifying road congestion (CN202411632935.1), this patent uses an image visual detection method to determine the degree of congestion by analyzing changes in the number of vehicles and similarity values. While this patent uses the same vehicle quantity detection method, it does not compare changes in the number of vehicles after detection. Instead, it uses the number of vehicles on a fixed road segment at a specific moment to represent the vehicle density and speed of that segment to determine the degree of road congestion.

[0004] Compared to the patented method, apparatus, computer equipment, and storage medium for assessing road congestion (CN202211714436.8), this patent identifies multiple road segments in a road network database, calculates the traffic flow of each segment, and combines the congestion index of each selected segment with its weighted contribution to the congestion status of all selected segments to calculate the congestion index of all selected segments, thus assessing the degree of road congestion. While similar to the vehicle quantity identification of selected road segments in this patent, it does not assess other indicators affecting congestion levels, such as vehicle speed, and it does not describe specific methods for detecting traffic flow on road segments.

[0005] Therefore, a strategy for assessing road congestion based on lidar detection is needed, which uses lidar to detect real-time vehicle data representing traffic flow on representative road segments, enabling real-time detection and updating of road congestion levels using lidar. Summary of the Invention

[0006] The purpose of this invention is to provide a method for assessing road congestion based on lidar detection, a device for assessing road congestion based on lidar detection, an electronic device, a storage medium, and a vehicle platform, thereby solving at least one of a number of technical problems.

[0007] For example, traditional computer vision channels are easily affected by light and occlusion when assessing congestion, and have limited recognition capabilities in outdoor scenes;

[0008] For example, some existing technologies rely solely on traffic flow as a single indicator to assess congestion, without incorporating key parameters such as vehicle speed, leading to inaccurate assessment results.

[0009] For example, some existing technologies do not clearly define the specific methods for detecting traffic flow and lack feasible technical solutions;

[0010] For example, existing technologies are insufficient to achieve real-time updates and quantitative classification of congestion status, and cannot meet the needs of drivers predicting routes and traffic management departments for dynamic control.

[0011] This invention provides the following solution:

[0012] According to a first aspect of the present invention, a method for assessing road congestion based on lidar detection is provided, comprising:

[0013] Step 1: Construct an ENU coordinate system with a preset reference point as the origin as the reference coordinate system, and calculate the external parameter relationship between the main lidar point cloud coordinate system and the reference coordinate system;

[0014] Step 2: On the XOY projection plane of the reference coordinate system, draw and record the closed boundary of the selected road segment;

[0015] Step 3: Preprocess the lidar point cloud, including coordinate transformation, removing point clouds outside the road segment boundaries and ground point clouds;

[0016] Step 4: Obtain the LiDAR vehicle detection method from deep learning;

[0017] Based on the LiDAR vehicle detection method in deep learning, all vehicles in a selected road segment at a certain time are detected and the number of vehicles is counted to obtain traffic flow data.

[0018] Step 5: By scanning the correlation and matching of vehicles in the previous and next frames, smoothing the vehicle trajectory with Kalman filter, calculating the real-time vehicle speed, and statistically analyzing the average and median speeds of all valid vehicles in the road segment;

[0019] Step 6: Based on traffic flow data, average vehicle speed, and median speed, obtain the congestion index through weighted analysis to determine the degree of congestion;

[0020] Step 7: Update and publish the road congestion assessment results at the preset frequency.

[0021] Furthermore, step 2 includes:

[0022] Obtain road segment boundary points through drone surveying, lidar surveying, or satellite map surveying;

[0023] The closed boundary is constructed using curve fitting.

[0024] Furthermore, step 3 includes:

[0025] Preprocessing of lidar point clouds:

[0026] The auxiliary lidar point cloud is converted to the main lidar coordinate system and stitched together. Then, all the obtained point clouds are converted to the ENU coordinate system.

[0027] Point clouds outside the road segment boundaries and above the preset height from the ground are removed, and the ground point cloud is segmented and removed using the Random Sample Consistency (RANSAC) algorithm.

[0028] Furthermore, step 4 includes:

[0029] The deep learning model is selected from Point R-CNN, Point Pillars, or SECOND models;

[0030] The vehicle detection process includes point cloud feature extraction, vehicle target prediction, confidence threshold filtering, and non-maximum suppression (NMS) processing to count the number of vehicles in the road segment.

[0031] Furthermore, step 5 includes:

[0032] The Hungarian algorithm or IOU matching method is used to achieve the correlation matching of vehicle frames before and after, and a unique ID is assigned to each vehicle;

[0033] The real-time speed of the vehicle is calculated by the displacement difference between the smooth positions of adjacent frames and the frame interval time, and abnormal speeds that exceed the reasonable range and vehicle speed data with a detection confidence of <0.7 are removed.

[0034] Where IOU ≥ 0.5.

[0035] Furthermore, step 6 includes:

[0036] First, the traffic flow, average speed, and median are standardized and mapped to the [0,1] interval.

[0037] The formula for calculating the congestion index using weighted analysis is: CI = α × (1 - V_avg_std) + β × (1 - V_med_std) + γ × F_std;

[0038] Where α+β+γ=1.0;

[0039] V_avg_std is the standardized value of the velocity average;

[0040] V_med_std is the standardized value of the median velocity;

[0041] F_std is the standardized value of traffic flow;

[0042] Among them, the weighting coefficients α, β, and γ are dynamically adjusted:

[0043] The default values ​​are α=0.3, β=0.3, and γ=0.4.

[0044] The values ​​for the highway section were adjusted to α=0.4, β=0.4, and γ=0.2.

[0045] The values ​​for urban intersection sections are adjusted to α=0.2, β=0.2, and γ=0.6.

[0046] The degree of congestion is classified according to the Congestion Index (CI):

[0047] CI < 0.2 indicates smooth traffic, 0.2 ≤ CI < 0.4 indicates mild congestion, 0.4 ≤ CI < 0.7 indicates moderate congestion, and CI ≥ 0.7 indicates severe congestion.

[0048] According to a second aspect of the present invention, a road congestion assessment device based on lidar detection is provided, which is based on a road congestion assessment method based on lidar detection, and includes:

[0049] The roadside includes a main lidar, several optional auxiliary lidars and their supports, as well as computer equipment and storage media for processing lidar point cloud information and performing detection functions.

[0050] Among them, the selection, assembly and installation position of the main lidar and the auxiliary lidar are designed to ensure that their radar field of view fully covers the selected road segment representing the degree of road congestion to be detected.

[0051] This also includes the fact that the lidar is selected from mechanical lidar, semi-solid-state lidar, or solid-state lidar;

[0052] Among them, after the auxiliary lidar is installed, its external parameters are calibrated with those of the main lidar to ensure unified processing of point cloud data.

[0053] According to a third aspect of the present invention, an electronic device is provided, comprising: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0054] The memory stores a computer program that, when executed by a processor, causes the processor to perform steps such as those in a road congestion assessment method based on lidar detection.

[0055] According to a fourth aspect of the present invention, a computer-readable storage medium is provided storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform steps such as a road congestion assessment method based on lidar detection.

[0056] According to a fifth aspect of the present invention, a vehicle platform is provided, comprising:

[0057] Electronic equipment for implementing steps in methods such as LiDAR-based road congestion assessment;

[0058] The processor runs programs that, when running, execute steps from data output by electronic devices, such as methods for assessing road congestion based on lidar detection.

[0059] Storage medium for storing programs that, when running, perform steps such as a road congestion assessment method based on LiDAR detection on data output from electronic devices.

[0060] The above solution achieves the following beneficial technical effects:

[0061] This application replaces cameras with LiDAR, which is unaffected by light or obstructions, and provides more stable vehicle detection in open outdoor scenes. It solves the environmental adaptability shortcomings of traditional vision solutions and significantly improves the anti-interference capability of road recognition.

[0062] This application combines three-dimensional indicators of "traffic flow + average vehicle speed + median speed" and uses a weighted algorithm to quantify the congestion index. Compared with single-indicator evaluation, it can more realistically reflect the congestion status of road segments and make the congestion assessment more accurate.

[0063] This application clarifies the complete detection process, including lidar selection, installation and calibration, point cloud preprocessing, and deep learning model application, thus addressing the deficiency of the comparative patent in "not describing the traffic flow detection method." It has practical engineering application value, and the detection solution is clear and feasible.

[0064] This application achieves congestion status updates every 0.5 seconds through high frame rate acquisition by lidar, Kalman filtering to smooth the trajectory, and iterative evaluation process. It can quickly respond to changes in traffic flow, meet the timeliness requirements of real-time navigation and traffic control, and has strong real-time and dynamic update capabilities.

[0065] In addition to ordinary roads, this application can be extended to logistics parks, parking lot entrances and exits, etc., filling the technical gap in traffic flow statistics and congestion assessment in specific closed / semi-closed areas, and has a wider range of applicable scenarios. Attached Figure Description

[0066] Figure 1 This is a flowchart of a road congestion assessment method based on lidar detection provided by one or more embodiments of the present invention.

[0067] Figure 2 This is a structural diagram of a road congestion assessment device based on lidar detection provided in one or more embodiments of the present invention.

[0068] Figure 3 This is a block diagram of an electronic device for assessing road congestion based on lidar detection, provided by one or more embodiments of the present invention. Detailed Implementation

[0069] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Figure 1 This is a flowchart of a road congestion assessment method based on lidar detection provided by one or more embodiments of the present invention.

[0071] like Figure 1 The road congestion assessment method based on lidar detection shown includes:

[0072] Step 1: Construct an ENU coordinate system with a preset reference point as the origin as the reference coordinate system, and calculate the external parameter relationship between the main lidar point cloud coordinate system and the reference coordinate system;

[0073] Step 2: On the XOY projection plane of the reference coordinate system, draw and record the closed boundary of the selected road segment;

[0074] Step 3: Preprocess the lidar point cloud, including coordinate transformation, removing point clouds outside the road segment boundaries and ground point clouds;

[0075] Step 4: Obtain the LiDAR vehicle detection method from deep learning;

[0076] Based on the LiDAR vehicle detection method in deep learning, all vehicles in a selected road segment at a certain time are detected and the number of vehicles is counted to obtain traffic flow data.

[0077] Step 5: By scanning the correlation and matching of vehicles in the previous and next frames, smoothing the vehicle trajectory with Kalman filter, calculating the real-time vehicle speed, and statistically analyzing the average and median speeds of all valid vehicles in the road segment;

[0078] Step 6: Based on traffic flow data, average vehicle speed, and median speed, obtain the congestion index through weighted analysis to determine the degree of congestion;

[0079] Step 7: Update and publish the road congestion assessment results at the preset frequency.

[0080] Specifically, the East-North-Sky coordinate system (ENU) is a three-dimensional rectangular coordinate system with a specific reference point as its origin. Its core is that the three axes are bound to the actual geographical location.

[0081] East (E-axis): Points to due east geographically;

[0082] North (N-axis): Points towards true north;

[0083] Sky (U-axis): Perpendicular to the ground, pointing towards the sky.

[0084] It can be compared to "standing at a reference point on the road section": facing east (E-axis), with north on the left (N-axis) and the sky directly above (U-axis). The three directions are perpendicular to each other, which can accurately locate the three-dimensional spatial position of vehicles within the road section, which is just right for the scenario of LiDAR roadside detection.

[0085] The XOY projection plane is a horizontal plane formed by "East (X-axis) - North (Y-axis)" in the ENU coordinate system (East-North-Sky). Essentially, it ignores the height dimension of "Sky (Z-axis)" and only retains the horizontal geographic plane, which is equivalent to "detecting the ground where the road segment is located".

[0086] This application uses it to draw road segment boundaries, which means "projecting" the geographical range of the road segment onto this horizontal plane, clearly defining the ground area to be detected (such as the horizontal outline of a certain road section), making it easier to subsequently filter vehicle point cloud data that are "only located within this ground area" and eliminate interference from high altitudes and underground.

[0087] In this embodiment, step 2 includes:

[0088] Obtain road segment boundary points through drone surveying, lidar surveying, or satellite map surveying;

[0089] The closed boundary is constructed using curve fitting.

[0090] In this embodiment, step 3 includes:

[0091] Preprocessing of lidar point clouds:

[0092] The auxiliary lidar point cloud is converted to the main lidar coordinate system and stitched together. Then, all the obtained point clouds are converted to the ENU coordinate system.

[0093] Point clouds outside the road segment boundaries and above the preset height from the ground are removed, and the ground point cloud is segmented and removed using the Random Sample Consistency (RANSAC) algorithm.

[0094] Specifically, the RANSAC (Random Sample Consistency) algorithm is a robust algorithm that accurately "picks out and removes ground points" from a messy LiDAR point cloud. Its core logic is "finding patterns and distinguishing between inside and outside": the ground point cloud approximately satisfies a "planar model". The algorithm fits a plane by randomly sampling a few points, selects the "inside points" (ground points) that conform to the plane, and removes the "outside points" (obstacles such as vehicles) that do not conform to the plane, and finally achieves the separation of ground and non-ground.

[0095] This can be likened to: viewing all the point clouds (ground + vehicles + noise) collected by the lidar as "pebbles (vehicles), dust (noise) and the ground itself scattered on a flat surface", and RANSAC as "using a flat plate for comparison" - randomly selecting 3 points (which can define a plane) and placing the flat plate on it. The parts that fit the flat plate are the ground, and the pebbles and dust that are lifted up are the non-ground. By repeatedly adjusting the flat plate to find the position that fits the ground best, the ground can be accurately "peeled" away.

[0096] Specific applications in this application (point cloud preprocessing scenarios)

[0097] Algorithm objective: In the point cloud preprocessing stage, remove ground point clouds and retain only the point clouds of obstacles such as vehicles to avoid ground points interfering with subsequent vehicle detection (e.g., not mistaking road surfaces for vehicles).

[0098] Key steps:

[0099] Assuming the ground is a "three-dimensional plane", we randomly sample three points from the point cloud (three points define a plane) and fit an initial plane model.

[0100] Calculate the distance from all remaining points to the plane, set a distance threshold (e.g., 0.1 meters), and determine points less than the threshold as "inner points" (which are likely ground points).

[0101] Repeat the iteration multiple times (e.g., 1000 times), resample the fitted plane each time, and finally select the plane with the "most interior points" as the "real ground model";

[0102] After removing all interior points (ground point cloud) corresponding to the model, what remains is the effective point cloud of obstacles such as vehicles.

[0103] Core technology effect

[0104] Strong anti-interference capability: It can ignore noise in point clouds (such as sensor error, small ground bumps) and a small number of abnormal points, accurately segment the ground, and solve the problem of "confusion between ground points and low-lying obstacle points";

[0105] Adaptable to complex road surfaces: Even if the ground has slight slopes or undulations (such as gentle slopes), it can still fit an approximate plane and will not fail to segment due to the road surface not being perfectly flat.

[0106] Paving the way for subsequent detection: After completely eliminating ground interference, the vehicle point cloud is purer, which can improve the accuracy of subsequent deep learning models in detecting vehicles (avoiding false detection of ground points as vehicles).

[0107] In this embodiment, step 4 includes:

[0108] The deep learning model is selected from Point R-CNN, Point Pillars, or SECOND models;

[0109] The vehicle detection process includes point cloud feature extraction, vehicle target prediction, confidence threshold filtering, and non-maximum suppression (NMS) processing to count the number of vehicles in the road segment.

[0110] Specifically, Non-maximum Suppression (NMS) is the key algorithm for "deduplication" after vehicle detection. The core logic is "keeping strong and deleting weak": the same vehicle may be detected by the model as multiple overlapping 3D bounding boxes. NMS will retain the box with the highest confidence (most reliable) and delete other "redundant boxes" with overlap exceeding the threshold, ensuring that each vehicle is counted only once.

[0111] This can be compared to taking multiple photos of the same car, some clear (high confidence) and some blurry (low confidence), with overlapping edges. NMS acts like a "filter"—first selecting the clearest photo as the "standard photo," then deleting all the blurry photos that overlap too much with it, ultimately keeping only the most reliable one to avoid duplicate statistics.

[0112] Specific applications in this application (vehicle inspection and post-processing scenarios).

[0113] When to use: After the deep learning model outputs vehicle candidate boxes (including the confidence score of each box), it is used to clean up duplicate detection results;

[0114] Key steps:

[0115] Step 1: Sort the candidate boxes in descending order of their confidence scores (the higher the confidence score, the more likely it is to be a real vehicle).

[0116] Step 2: Take the first sorted box as the "base box" and keep it (it will be included in the traffic flow calculation later).

[0117] Step 3: Calculate the IOU (Intersection over Union) between all other boxes and the reference box. If the IOU is ≥ 0.7 (the document sets a threshold), it is determined to be an "overlapping redundant box" and is deleted directly.

[0118] Step 4: From the remaining boxes, select the one with the highest confidence as the new baseline box, and repeat step 3 until all boxes are determined to be "keep" or "delete".

[0119] Core objective: To avoid counting the same vehicle repeatedly in multiple boxes and to ensure the accuracy of traffic flow statistics (e.g., not counting the same vehicle as 2 or more vehicles).

[0120] Core technology effect

[0121] Accurate deduplication: Overlap detection is controlled by the IOU threshold, which not only deletes redundant duplicate boxes, but also avoids accidentally deleting adjacent boxes of different vehicles (such as two vehicles driving side by side).

[0122] Improve detection purity: Filter out low-confidence false positives (such as those that mistakenly identify roadside debris as vehicles), resulting in cleaner final vehicle detection results;

[0123] Ensuring the reliability of subsequent processes: Accurate traffic flow statistics are the foundation for subsequent vehicle speed calculations and congestion index assessments, and NMS provides a safety net for the accuracy of the entire congestion assessment chain.

[0124] In this embodiment, step 5 includes:

[0125] The Hungarian algorithm or IOU matching method is used to achieve the correlation matching of vehicle frames before and after, and a unique ID is assigned to each vehicle;

[0126] The real-time speed of the vehicle is calculated by the displacement difference between the smooth positions of adjacent frames and the frame interval time, and abnormal speeds that exceed the reasonable range and vehicle speed data with a detection confidence of <0.7 are removed.

[0127] Where IOU ≥ 0.5.

[0128] Specifically, the IOU (Intersection over Union) matching method is an algorithm for determining the degree of overlap between two detection boxes (in this application, the vehicle's three-dimensional bounding box). The core of it is to calculate the ratio of the "area of ​​the overlapping region between the two boxes" to the "area of ​​the total coverage region between the two boxes" (the value ranges from 0 to 1).

[0129] This can be compared to two vehicle detection frames being like two pieces of paper. The IOU is the area of ​​the two frames stacked together, divided by the total area of ​​the two frames (minus the overlapping part to avoid double counting). The closer the ratio is to 1, the more overlap there is.

[0130] In this application, it is used to achieve vehicle frame association matching: if the IOU of the vehicle's 3D bounding box is ≥0.5 in two consecutive frames (such as frame t and frame t-1), it is determined to be the same vehicle, and a unique ID is assigned to ensure the accuracy of subsequent trajectory tracking and vehicle speed calculation. It is suitable for normal scenarios where traffic flow is not extremely dense, and is simple, efficient and easy to implement.

[0131] The Hungarian algorithm is a classic algorithm for solving "many-to-many optimal matching". Its core is to achieve one-to-one accurate matching by minimizing the total "cost" (such as distance and difference) in two sets of objects (vehicle detection boxes in two consecutive frames in this application), thus avoiding duplicate matching or missed matching.

[0132] This can be likened to a "vehicle matching game": all vehicles in frame t are considered "players to be matched", and all vehicles in frame t-1 are considered "target players". The "distance" between each pair of players (such as the Euclidean distance between the center points of the vehicles) is the "matching cost". The algorithm will find a matching scheme with the minimum total cost to ensure that each vehicle is matched with only one corresponding vehicle from the previous frame (or is determined to be newly appeared / already left).

[0133] In this application, it is used to achieve vehicle frame-to-frame correlation matching:

[0134] For scenarios with dense traffic (such as multiple vehicles driving side by side in severe congestion), avoid the "mismatch caused by multiple overlapping vehicles" problem that may occur with the IOU matching method;

[0135] Using the Euclidean distance between the center points of the vehicle's 3D bounding box as the cost function, the algorithm calculates the optimal matching relationship and assigns a unique ID to each vehicle to ensure the continuity of subsequent trajectory tracking and vehicle speed calculation (for example, it will not misclassify two adjacent vehicles as the same vehicle).

[0136] It complements the IOU matching method: IOU is suitable for regular traffic flow (simple and efficient), while the Hungarian algorithm is suitable for dense traffic flow (more accurate matching), together ensuring the reliability of cross-frame association.

[0137] In this embodiment, step 6 includes:

[0138] First, the traffic flow, average speed, and median are standardized and mapped to the [0,1] interval.

[0139] The formula for calculating the congestion index using weighted analysis is: CI = α × (1 - V_avg_std) + β × (1 - V_med_std) + γ × F_std;

[0140] Where α+β+γ=1.0;

[0141] V_avg_std is the standardized value of the velocity average;

[0142] V_med_std is the standardized value of the median velocity;

[0143] F_std is the standardized value of traffic flow;

[0144] Among them, the weighting coefficients α, β, and γ are dynamically adjusted:

[0145] The default values ​​are α=0.3, β=0.3, and γ=0.4.

[0146] The values ​​for the highway section were adjusted to α=0.4, β=0.4, and γ=0.2.

[0147] The values ​​for urban intersection sections are adjusted to α=0.2, β=0.2, and γ=0.6.

[0148] The degree of congestion is classified according to the Congestion Index (CI):

[0149] CI < 0.2 indicates smooth traffic, 0.2 ≤ CI < 0.4 indicates mild congestion, 0.4 ≤ CI < 0.7 indicates moderate congestion, and CI ≥ 0.7 indicates severe congestion.

[0150] Figure 2 This is a structural diagram of a road congestion assessment device based on lidar detection provided in one or more embodiments of the present invention.

[0151] like Figure 2 The road congestion assessment device based on lidar detection shown is based on a road congestion assessment method based on lidar detection. The road congestion assessment device based on lidar detection includes:

[0152] The roadside includes a main lidar, several optional auxiliary lidars and their supports, as well as computer equipment and storage media for processing lidar point cloud information and performing detection functions.

[0153] Among them, the selection, assembly and installation position of the main lidar and the auxiliary lidar are designed to ensure that their radar field of view fully covers the selected road segment representing the degree of road congestion to be detected.

[0154] This also includes the fact that the lidar is selected from mechanical lidar, semi-solid-state lidar, or solid-state lidar;

[0155] Among them, after the auxiliary lidar is installed, its external parameters are calibrated with those of the main lidar to ensure unified processing of point cloud data.

[0156] It is worth noting that although this system / device only discloses the above-mentioned modules / units, it does not mean that this system / device is limited to the above-mentioned basic functional modules. On the contrary, what this invention intends to express is that, based on the above-mentioned basic functional modules, those skilled in the art can add one or more functional modules in combination with the prior art to form an infinite number of embodiments or technical solutions. That is to say, this system is open rather than closed. It cannot be assumed that the scope of protection of the claims of this invention is limited to the above-disclosed basic functional modules just because this embodiment only discloses a few basic functional modules.

[0157] In one specific embodiment, a road congestion assessment strategy based on lidar detection is disclosed, which is used to detect and update road congestion levels in real time using lidar.

[0158] In this embodiment, a road congestion detection device is first deployed, including at least one main lidar; or several auxiliary lidars and brackets are added and set at the roadside end, which are computer equipment and media used to process lidar point cloud information and complete the detection function.

[0159] The selection, assembly, and installation locations of the main and auxiliary lidars should ensure that the radar's field of view can cover the selected road segment that represents the degree of road congestion.

[0160] In this embodiment, the method for assessing road congestion using lidar detection includes constructing a three-dimensional spatial coordinate system, drawing the regional boundaries of selected road segments, preprocessing lidar point clouds, detecting traffic flow, detecting vehicle speed, analyzing congestion, and updating and publishing congestion information.

[0161] Among them, the construction of the three-dimensional spatial coordinate system includes the construction of a reference coordinate system, that is, to establish an ENU coordinate system (East-North-Sky coordinate system) with a specific reference point as the origin.

[0162] Drawing the regional boundary of the selected road segment includes drawing and recording the closed boundary of the selected road segment in the main lidar coordinate system constructed above.

[0163] The lidar point cloud preprocessing includes lidar point cloud coordinate transformation, removing lidar point clouds that do not belong to the selected road segment boundary drawn above, and then removing ground points.

[0164] Traffic flow detection includes using the pre-processed radar point cloud and the LiDAR vehicle detection method in deep learning to detect all vehicles in the selected road segment at a certain time and count the number of vehicles in the selected road segment at that time.

[0165] The vehicle speed detection process includes matching the detected vehicles with consecutive frames, smoothing the trajectory of each vehicle using Kalman filtering, calculating the real-time speed of the vehicles using time difference, and statistically calculating the average and median speeds of all vehicles passing through the road segment to assess the degree of congestion.

[0166] The congestion analysis includes using a weighted analysis based on the average and median speeds of traffic flow and passing vehicles to provide feedback on the congestion level of the road segment.

[0167] The congestion update and publication process includes updating and publishing the congestion status of the road segment in real time based on the congestion assessment results from the previous step.

[0168] In this embodiment, regarding the selection, assembly, and installation of the lidar:

[0169] Depending on the size and shape of the selected road segment, a primary lidar unit or a primary lidar unit and several auxiliary lidar units can be selected. Lidar units can be mechanical, semi-solid-state, or solid-state. The lidar units are fixed to the roadside using equipment poles or similar means, and their angles are adjusted. Due to the varying road segments, the installation principle is that the configured lidar field of view should cover the selected road segment. After installation, the extrinsic parameters of the auxiliary and primary lidar units are calibrated for subsequent processing.

[0170] In this embodiment, regarding the assessment and updating of road congestion levels:

[0171] Step 1: Constructing a 3D spatial coordinate system. Select a suitable point as the origin and establish an ENU coordinate system, i.e., the East-North-Sky coordinate system, as the reference system for subsequent point cloud analysis and drawing of selected road segment boundaries. Calculate the extrinsic parameters between the main lidar point cloud coordinate system and this reference coordinate system.

[0172] Step Two: Draw the boundary of the selected road segment. Using the coordinate system established in Step One, draw the boundary of the selected road segment on the XOY projection plane of the coordinate system through methods such as UAV surveying, LiDAR surveying, and satellite mapping. Measure several boundary points and construct the closed boundary of the road segment through curve fitting.

[0173] Step 3: LiDAR Point Cloud Preprocessing. First, based on the extrinsic parameter relationship between the LiDAR and its coordinate system, the coordinates of the auxiliary LiDAR point cloud are transformed to the coordinate system of the main LiDAR point cloud through translation and rotation, and the point clouds are stitched together to form a complete point cloud image covering the entire LiDAR field of view. Then, the coordinates of the entire LiDAR point cloud are transformed to the ENU reference coordinate system through translation and rotation. Next, all point clouds are filtered, deleting those not within the road segment boundaries drawn in Step 2, and deleting those too high above the parking lot ground. Finally, the ground point cloud is segmented using the Random Sample Consensus (RANSAC) algorithm or other algorithms and then deleted. The final result is the preprocessed complete point cloud of the road segment.

[0174] Step 4: Obstacle Detection. A dedicated detection model is constructed using deep learning-based LiDAR vehicle detection methods, specifically including:

[0175] 1. Construction and training of the LiDAR vehicle detection model:

[0176] 1) Model architecture selection: Based on the sparsity and 3D spatial features of LiDAR point clouds, a deep learning model suitable for point cloud detection is selected, such as:

[0177] Models based on direct point cloud processing (such as PointRCNN, PointPillars, SECOND, etc.) can achieve 3D bounding box detection of vehicle targets by converting point clouds into pseudo-images (such as voxelization, cylindrical feature encoding) or by directly learning features from the original points.

[0178] 2) Training data collection: In road environments similar to the target road segment (such as urban main roads, highways, etc.), collect LiDAR point cloud data under different time periods (day / night) and different weather conditions (sunny / rainy / foggy), and manually annotate the vehicle targets in the data. The annotation content includes the vehicle's 3D bounding box (center point coordinates, length / width / height, orientation angle) and category (such as car, truck, bus, etc., optional).

[0179] 3) Data augmentation: Enhance the training data, including point cloud rotation, scaling, noise addition, and cropping, to improve the model's robustness to different vehicle postures, distances, and interference scenarios;

[0180] 4) Training process: With the positioning accuracy (such as IoU index) and classification accuracy of the vehicle 3D bounding box as the optimization goal, stochastic gradient descent (SGD) or Adam optimizer is used to train the model through multiple rounds of iterations to achieve convergence and save the optimal model weights.

[0181] 2. The reasoning process for vehicle inspection:

[0182] Input the preprocessed point cloud frame into the trained detection model and perform the following inference steps:

[0183] 1) Point cloud feature extraction: The model performs hierarchical feature learning on the input point cloud, for example:

[0184] 2) For the PointPillars model: First, the point cloud is divided into pillars. Then, the features of each pillar are extracted using PillarFeatureNet and converted into a pseudo-image feature map.

[0185] 3) For the PointRCNN model: first, global and local features of the point cloud are extracted through PointNet to generate candidate vehicle boxes, and then the position and size of the boxes are optimized through RoI pooling.

[0186] 4) Vehicle target prediction: Based on the extracted features, the model outputs detection results, including:

[0187] The 3D bounding box parameters of each candidate vehicle (used to determine the spatial location of the vehicle in the road segment);

[0188] Confidence score (used to determine the probability that the candidate box is a real vehicle; for example, setting the confidence threshold to 0.5 will filter out false positives with low confidence).

[0189] 5) Post-processing optimization: Non-maximum suppression (NMS) is applied to the candidate boxes output by the model to remove duplicate detection boxes of the same vehicle (e.g., the box with the highest confidence is retained when IoU is greater than 0.7), ensuring that each vehicle is detected only once.

[0190] 3. Statistics on the number of vehicles within the selected road section:

[0191] Based on the detection results, the number of vehicles in a selected road segment at a certain time is counted, according to the following rules:

[0192] 1) Spatial filtering: Based on the preset selected road segment boundary, determine whether the center point of the detected vehicle's 3D bounding box is within the road segment range, and only count vehicles within the road segment;

[0193] 2) Deduplication verification: If the same vehicle is detected in consecutive frames (associated by timestamp), but this step only applies to the single-frame point cloud at "a certain moment", there is no need to deduplicat across frames, only to ensure that there are no duplicate counts within a single frame.

[0194] 3) Output the number of vehicle detection boxes that meet the above conditions. This number is used as the traffic flow data for the selected road segment at that time. The output format is an integer (e.g., "N vehicles are in the selected road segment at that time").

[0195] Step 5: Vehicle speed detection.

[0196] 1. Regarding the matching of vehicle frames before and after:

[0197] For point cloud frames with consecutive timestamps (where the time interval between adjacent frames is Δt, determined by the lidar frame rate, e.g., Δt = 0.1 seconds for a 10Hz lidar), cross-frame association is performed on detected vehicle targets to ensure that the same vehicle is uniquely identified in different frames. Specific steps are as follows:

[0198] Regarding vehicle feature extraction: For each vehicle detected in the current frame (frame t) and the previous frame (frame t-1), features for association are extracted, including:

[0199] Regarding spatial location features: coordinates of the center point of the vehicle's 3D bounding box (x... t ,y t ,z t (current frame) and (x) t ₋1,y t ₋1,z t ₋1)(Previous frame);

[0200] Regarding morphological features: the length, width, and height of the vehicle's bounding box (used to distinguish different types of vehicles, such as trucks and cars).

[0201] Regarding motion trend features: the current frame position predicted based on historical trajectories (if any) (applicable to multi-frame associated scenarios).

[0202] Regarding the association matching algorithm: the Hungarian Algorithm or the IOU (Intersection over Union) matching method is used to implement vehicle association.

[0203] If IOU matching is used: calculate the IOU of the 3D bounding boxes of vehicle i in frame t and vehicle j in frame (t-1). When IOU ≥ 0.5, they are determined to be the same vehicle.

[0204] Regarding the use of the Hungarian algorithm: the Euclidean distance between the center points of vehicles in two frames is used as the cost function, and the optimal matching is achieved by minimizing the total cost (suitable for dense vehicle scenes, reducing false matching).

[0205] Regarding unique identifier allocation: A unique ID (such as "Vehicle_001") is assigned to a successfully associated vehicle. If a vehicle appears for the first time in the current frame (without a preceding match), a new ID is created. If a vehicle is not matched in the current frame (e.g., it has left the road segment), it is marked as "left" and its historical trajectory is retained.

[0206] 2. Regarding vehicle trajectory smoothing based on Kalman filtering:

[0207] A Kalman filter is applied to the continuous position sequence of each vehicle (sorted by timestamp) to eliminate trajectory jitter caused by detection noise, and a smoothed trajectory is output. Specific implementation details are as follows:

[0208] 1) State model construction:

[0209] State vector: Define the vehicle state as [position x, position y, velocity v] x [Velocity vᵧ] (two-dimensional planar motion, ignoring vertical motion in the z direction), where position is the coordinate in the lidar coordinate system, and velocity is the instantaneous velocity in the x and y directions;

[0210] Regarding the state transition equation: Based on the uniform motion model, assuming the vehicle speed remains constant between adjacent frames, the state transition matrix is:

[0211] ;

[0212] (Δt is the frame interval time, in seconds).

[0213] 2) Construction of the observation model:

[0214] Observation vector: The coordinates (x_obs, y_obs) of the center point of the bounding box obtained from vehicle detection are used as the observation values;

[0215] Observation matrix: This matrix maps the positional information in the state vector to the observation space.

[0216] ;

[0217] 3) Filtering iteration process:

[0218] Prediction: Based on the smoothed state and state transition equation of the previous time step, predict the vehicle state at the current time step;

[0219] Update: Combining the detection position (observation value) of the current frame, the predicted state is adjusted by Kalman gain to obtain the smoothed state of the current time step (including smoothed position and velocity).

[0220] Initial value setting: When the vehicle first appears, the detection position is used as the initial position, and the initial speed is set to 0 or estimated based on the difference between the detection positions of the previous two frames.

[0221] Output: After processing each frame, output the smooth trajectory points (x_smooth, y_smooth) of the vehicle and the corresponding timestamps for subsequent speed calculations.

[0222] 3. Real-time vehicle speed calculation:

[0223] Based on the smoothed trajectory, the real-time speed of each vehicle is calculated using the time difference. The specific method is as follows:

[0224] 1) Instantaneous velocity calculation:

[0225] The smoothed position (x) of a vehicle in frame t and frame (t-1) t ,y t ) and (x t ₋1,y t ₋1), Calculate the two-dimensional displacement distance:

[0226] Displacement d = √[(x t -x t ₋1)²+(y t -y t ₋1)²]

[0227] Combining the frame interval time Δt, the instantaneous velocity is obtained:

[0228] vᵢ nst =d / Δt (unit: meters per second, can be converted to kilometers per hour: v(km / h)=vᵢ) nst ×3.6)

[0229] 2) Speed ​​validity verification:

[0230] Outlier removal: If the calculated instantaneous speed exceeds the reasonable range (e.g., urban roads ≤ 60km / h, highways ≤ 120km / h), the speed status output by the Kalman filter is used instead.

[0231] Low confidence filtering: For vehicles with a detection confidence level < 0.7, their speed results are marked as "unreliable" and are not included in subsequent statistics.

[0232] 4. Vehicle speed statistics for the road section:

[0233] The speeds of all valid vehicles within the selected road segment are statistically analyzed, and the average and median speeds are calculated as indicators of congestion levels. Specific steps include:

[0234] 1) Definition of statistical scope:

[0235] Time range: Centered on a "certain moment", take velocity data for N consecutive frames (e.g., N=5, covering 0.5 seconds) to avoid the influence of single frame fluctuations;

[0236] Spatial range: Only vehicles currently located within the selected road segment are counted (determined by whether the smooth position of the vehicle is within the geographical boundary of the road segment).

[0237] 2) Calculation of statistics:

[0238] Average speed: the arithmetic mean of the speeds of all valid vehicles, formula:

[0239] v_avg=(v1+v2+...+v k) / k (k is the number of valid vehicles)

[0240] Median speed: Sort all valid vehicle speeds in ascending order and take the median value (if k is even, take the average of the two median values), denoted as v_med.

[0241] Step Six: Congestion Analysis. Congestion analysis is based on the acquired traffic flow data, average vehicle speed, and median speed of the selected road segment. These three factors are weighted using preset weights to calculate a quantified congestion index, thereby determining the degree of congestion on the road segment. The specific steps are as follows:

[0242] 1. Standardization of basic parameters:

[0243] To eliminate the dimensional differences between traffic flow and speed parameters (traffic flow is measured in "vehicles / unit time", and speed is measured in "km / h"), the original data needs to be standardized and mapped uniformly to the [0,1] interval. The specific method is as follows:

[0244] 1) Traffic flow standardization (F_std):

[0245] Define the maximum saturated traffic flow of a road segment as F_max (unit: vehicles / hour), which is the maximum capacity of the road segment under ideal conditions (it can be preset by parameters such as the number of lanes and design speed of the road segment, such as F_max = 2000 vehicles / hour for a two-way four-lane main road).

[0246] 2) Let the measured traffic flow at a certain moment be F (vehicles / hour), then the standardized traffic flow is:

[0247] F_std=min(F / F_max,1.0)

[0248] (When F≥F_max, F_std=1.0, which means the traffic flow has reached saturation).

[0249] 3) Velocity average standardization (V_avg_std):

[0250] Define the free-flow speed V_free (unit: km / h), which is the average driving speed under no-congestion conditions (e.g., V_free = 50 km / h on urban arterial roads).

[0251] 4) Let the average measured speed at a certain moment be V_avg (km / h), then the standardized speed mean is:

[0252] V_avg_std=max(V_avg / V_free,0.0)

[0253] (When V_avg=0, V_avg_std=0; when V_avg≥V_free, V_avg_std=1.0, representing no speed loss).

[0254] 5) Median velocity standardization (V_med_std):

[0255] Standardizing using the same free-flow velocity V_free as the velocity mean, and assuming the median measured velocity is V_med (km / h), then:

[0256] V_med_std=max(V_med / V_free,0.0)

[0257] 2. Construction and parameter setting of the weighted analysis model:

[0258] Based on standardized parameters, a weighted congestion index model is constructed. The weight allocation reflects the influence of traffic flow, average speed, and median speed on congestion assessment. Specifically:

[0259] 1) Congestion Index (CI) Calculation Formula:

[0260] CI=α×(1-V_avg_std)+β×(1-V_med_std)+γ×F_std

[0261] in:

[0262] (1-V_avg_std) and (1-V_med_std) are "velocity loss factors": the larger the value, the further the velocity deviates from the free flow, and the more obvious the congestion tendency.

[0263] F_std is the "traffic saturation factor": the larger the value, the closer the traffic flow is to saturation, and the higher the risk of congestion.

[0264] α, β, and γ are weighting coefficients that satisfy α+β+γ=1.0 and are dynamically adjusted according to road segment characteristics (such as whether it is a bottleneck segment and traffic flow stability). The default values ​​are: α=0.3, β=0.3, γ=0.4 (prioritizing the impact of traffic flow on congestion, while supplementing the stability of speed distribution by the median speed).

[0265] 2) Weighting adjustment rules:

[0266] If the road segment is a highway (traffic flow is mainly at a constant speed and the speed distribution is concentrated), the weight of the speed parameter can be increased: α=0.4, β=0.4, γ=0.2;

[0267] If the road segment is near an urban intersection (where traffic flow fluctuates greatly and is prone to saturation), the traffic flow weight can be increased: α=0.2, β=0.2, γ=0.6.

[0268] 2. Classification of congestion levels:

[0269] Based on the calculated congestion index CI (with a value range of [0,1]), the congestion level of road segments is divided into 4 levels, with specific grading criteria as follows:

[0270] 1) Unobstructed (CI < 0.2):

[0271] Judgment criteria: Traffic flow is far below the saturation value (F_std < 0.3), and the average and median speeds are close to free flow (V_avg_std ≥ 0.8, V_med_std ≥ 0.8).

[0272] Typical characteristics: Vehicles move smoothly without significant deceleration or queuing.

[0273] 2) Mild congestion (0.2 ≤ CI < 0.4):

[0274] Judgment criteria: moderate traffic volume (0.3≤F_std<0.5), slight speed loss (0.6≤V_avg_std<0.8, 0.6≤V_med_std<0.8);

[0275] Typical characteristics: Some vehicles slow down, and traffic begins to slow down intermittently.

[0276] 3) Moderate congestion (0.4 ≤ CI < 0.7):

[0277] Judgment criteria: Traffic flow is close to saturation (0.5≤F_std<0.8), and speed has decreased significantly (0.3≤V_avg_std<0.6, 0.3≤V_med_std<0.6).

[0278] Typical characteristics: Most vehicles are traveling at low speeds, resulting in continuous queues with a queue length of less than 500 meters.

[0279] 4) Severe congestion (CI≥0.7):

[0280] Judgment criteria: Traffic flow reaches or exceeds saturation (F_std≥0.8), and speed decreases significantly (V_avg_std<0.3, V_med_std<0.3);

[0281] Typical characteristics: vehicles are stationary or moving at a snail's pace (≤10km / h), queue length is ≥500 meters, and traffic efficiency is significantly reduced.

[0282] 3. Feedback and application of congestion analysis results:

[0283] The congestion level classification results will be output in a visual or data interface format to support subsequent applications. Specifically:

[0284] 1) Output format:

[0285] Quantitative indicators: Directly output the congestion index CI (retaining 2 decimal places) and the corresponding classification label (such as "severe congestion");

[0286] Auxiliary information: Synchronously output raw parameters (traffic flow F, average speed V_avg, median speed V_med) for retrospective analysis;

[0287] Dynamic update frequency: consistent with the time granularity of traffic flow and speed detection (e.g., updated every 0.5 seconds) to ensure real-time reflection of changes in road congestion.

[0288] Step 7: Update and publish congestion levels. The above steps can be executed cyclically at a certain frequency, and the congestion level assessment results of each execution will be updated and published.

[0289] Figure 3 This is a block diagram of an electronic device for assessing road congestion based on lidar detection, provided by one or more embodiments of the present invention.

[0290] like Figure 3 As shown, this application provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0291] The memory stores a computer program that, when executed by the processor, causes the processor to perform steps of a road congestion assessment method based on lidar detection.

[0292] This application also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a road congestion assessment method based on lidar detection.

[0293] This application also provides a vehicle platform, including:

[0294] Electronic equipment for implementing steps in a method for assessing road congestion based on lidar detection;

[0295] The processor runs a program that, when running, executes steps of a road congestion assessment method based on LiDAR detection from data output by electronic devices.

[0296] Storage medium for storing programs that, when running, execute steps of a road congestion assessment method based on lidar detection on data output from electronic devices.

[0297] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0298] The electronic device comprises a hardware layer, an operating system layer running on top of the hardware layer, and an application layer running on the operating system. The hardware layer includes hardware such as a central processing unit (CPU), a memory management unit (MMU), and memory. The operating system can be any one or more computer operating systems that control the electronic device through processes, such as Linux, Unix, Android, iOS, or Windows. Furthermore, in this embodiment of the invention, the electronic device can be a smartphone, tablet computer, or other handheld device, or a desktop computer, portable computer, or other electronic device; there is no particular limitation in this embodiment.

[0299] In this embodiment of the invention, the executing entity for electronic device control can be an electronic device itself, or a functional module within an electronic device capable of calling and executing a program. The electronic device can obtain the firmware corresponding to the storage medium. This firmware is provided by the supplier, and different storage media may have the same or different firmware; no limitation is made here. After obtaining the firmware corresponding to the storage medium, the electronic device can write this firmware into the storage medium; specifically, it burns the firmware corresponding to the storage medium into the storage medium. The process of burning the firmware into the storage medium can be implemented using existing technology, and will not be elaborated upon in this embodiment of the invention.

[0300] Electronic devices can also obtain reset commands corresponding to storage media. These reset commands are provided by the supplier, and the reset commands for different storage media can be the same or different, which is not limited here.

[0301] At this time, the storage medium of the electronic device is a storage medium on which the corresponding firmware has been written. The electronic device can respond to the reset command corresponding to the storage medium on which the corresponding firmware has been written, thereby resetting the storage medium on which the corresponding firmware has been written according to the reset command. The process of resetting the storage medium according to the reset command can be implemented by existing technology and will not be described in detail in this embodiment of the invention.

[0302] For ease of description, the above devices are described separately by function as various units and modules. Of course, in implementing this application, the functions of each unit and module can be implemented in one or more software and / or hardware.

[0303] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined.

[0304] For the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0305] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0306] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing road congestion based on lidar detection, characterized in that, include: Step 1: Construct an ENU coordinate system with a preset reference point as the origin as the reference coordinate system, and calculate the external parameter relationship between the main lidar point cloud coordinate system and the reference coordinate system; Step 2: On the XOY projection plane of the reference coordinate system, draw and record the closed boundary of the selected road segment; Step 3: Preprocess the lidar point cloud, including coordinate transformation, removing point clouds outside the road segment boundaries and ground point clouds; Step 4: Obtain the LiDAR vehicle detection method from deep learning; Based on the LiDAR vehicle detection method in deep learning, all vehicles in a selected road segment at a certain time are detected and the number of vehicles is counted to obtain traffic flow data. Step 5: By scanning the correlation and matching of vehicles in the previous and next frames, smoothing the vehicle trajectory with Kalman filter, calculating the real-time vehicle speed, and statistically analyzing the average and median speeds of all valid vehicles in the road segment; Step 6: Based on the traffic flow data, average vehicle speed, and median speed, obtain the congestion index through weighted analysis to determine the degree of congestion; Step 7: Update and publish the road congestion assessment results at the preset frequency.

2. The method for assessing road congestion based on lidar detection according to claim 1, characterized in that, Step 2 includes: Obtain road segment boundary points through drone surveying, lidar surveying, or satellite map surveying; The closed boundary is constructed using curve fitting.

3. The method for assessing road congestion based on lidar detection according to claim 1, characterized in that, Step 3 includes: Preprocessing of lidar point clouds: The auxiliary lidar point cloud is converted to the main lidar coordinate system and stitched together. Then, all the obtained point clouds are converted to the ENU coordinate system. Point clouds outside the road segment boundaries and above the preset height from the ground are removed, and the ground point cloud is segmented and removed using the Random Sample Consistency (RANSAC) algorithm.

4. The method for assessing road congestion based on lidar detection according to claim 1, characterized in that, Step 4 includes: The deep learning model is selected from Point R-CNN, Point Pillars, or SECOND models; The vehicle detection process includes point cloud feature extraction, vehicle target prediction, confidence threshold filtering, and non-maximum suppression (NMS) processing to count the number of vehicles in the road segment.

5. The method for assessing road congestion based on lidar detection according to claim 1, characterized in that, Step 5 includes: The Hungarian algorithm or IOU matching method is used to achieve the correlation matching of vehicle frames before and after, and a unique ID is assigned to each vehicle; The real-time speed of the vehicle is calculated by the displacement difference between the smooth positions of adjacent frames and the frame interval time, and abnormal speeds that exceed the reasonable range and vehicle speed data with a detection confidence of <0.7 are removed. Where IOU ≥ 0.

5.

6. The method for assessing road congestion based on lidar detection according to claim 3, characterized in that, Step 6 includes: First, the traffic flow, average speed, and median are standardized and mapped to the [0,1] interval. The formula for calculating the congestion index using weighted analysis is: CI = α × (1 - V_avg_std) + β × (1 - V_med_std) + γ × F_std; Where α+β+γ=1.0; V_avg_std is the standardized value of the velocity average; V_med_std is the standardized value of the median velocity; F_std is the standardized value of traffic flow; The weighting coefficients α, β, and γ are dynamically adjusted. The default values ​​are α=0.3, β=0.3, and γ=0.

4. The values ​​for the highway section were adjusted to α=0.4, β=0.4, and γ=0.

2. The values ​​for urban intersection sections are adjusted to α=0.2, β=0.2, and γ=0.

6. The degree of congestion is classified according to the Congestion Index (CI): CI < 0.2 indicates smooth traffic, 0.2 ≤ CI < 0.4 indicates mild congestion, 0.4 ≤ CI < 0.7 indicates moderate congestion, and CI ≥ 0.7 indicates severe congestion.

7. A road congestion assessment device based on lidar detection, characterized in that, The road congestion assessment method based on lidar detection according to any one of claims 1 to 6, wherein the road congestion assessment device based on lidar detection comprises: The roadside includes a main lidar, several optional auxiliary lidars and their supports, as well as computer equipment and storage media for processing lidar point cloud information and performing detection functions. The selection, assembly, and installation location of the main lidar and the auxiliary lidar are such that their radar field of view completely covers the selected road segment representing the degree of road congestion to be detected. It also includes the fact that the lidar is selected from mechanical lidar, semi-solid-state lidar or solid-state lidar; Among them, after the auxiliary lidar is installed, its external parameters are calibrated with those of the main lidar to ensure unified processing of point cloud data.

8. An electronic device, characterized in that, include: The processor, communication interface, memory, and communication bus are connected, with the processor, communication interface, and memory communicating with each other via the communication bus. The memory stores a computer program that, when executed by a processor, causes the processor to perform the steps of the road congestion assessment method based on lidar detection as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The device stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the road congestion assessment method based on lidar detection as described in any one of claims 1 to 6.

10. A vehicle platform, characterized in that, include: An electronic device for implementing the steps of the road congestion assessment method based on lidar detection as described in any one of claims 1 to 6; The processor runs a program that, when the program is running, performs the steps of the road congestion assessment method based on lidar detection as described in any one of claims 1 to 6 from data output by the electronic device. A storage medium for storing a program that, when running, performs the steps of the road congestion assessment method based on lidar detection as described in any one of claims 1 to 6 on data output from an electronic device.

Images