A retroreflector detection method and system based on laser radar

Abstract

The application discloses a retroreflector detection method and system based on a laser radar. The method specifically comprises the following steps: S1, extracting reflection point information of a laser radar signal; S2, converting and preprocessing a target point cloud; S3, extracting effective retroreflector coordinates based on distance gradient weights; S4, retroreflector feature identification; S5, abnormal retroreflector feature identification; and S6, alarming and recording. The application determines the retroreflector attribute by establishing retroreflector feature information through the laser radar, provides additional information to improve positioning accuracy, and reduces the difficulty of field test of a positioning algorithm.

Description

Technical Field

[0001] This invention relates to the field of lidar positioning technology, and in particular to a lidar-based method and system for detecting reflective objects. Background Technology

[0002] With the development of modern technology, lidar has been widely used in many fields, especially in environmental monitoring, autonomous driving, and 3D reconstruction. LiDAR obtains the 3D information of a target by sending a laser beam and receiving the reflected signal. However, due to the presence of various objects and interference factors in the environment, lidar signals often reflect back a lot of useless information, which brings great difficulties to target detection and recognition.

[0003] Existing single-line lidar reflector localization schemes rely on matching the center coordinates of selected reflectors to obtain localization information. The identification of reflectors depends on the reflection intensity of objects on the lidar scanning lines. Due to the single data source, it is impossible to accurately identify irrelevant objects, reflectors, and interfering reflectors, thus affecting the final localization result. Summary of the Invention

[0004] Based on this, embodiments of this application provide a method and system for detecting reflective objects based on lidar. The method uses lidar to establish reflective object feature information to determine the reflective object's attributes, including category, coordinates, range of associated points, and positioning reliability. This application improves positioning accuracy and reduces the difficulty of field testing the positioning algorithm by providing additional information.

[0005] Firstly, a method for detecting reflective objects based on lidar is provided, the method comprising:

[0006] S1, Extracting reflection point information from lidar signals: Obtaining reflection signals in the environment through lidar equipment and extracting reflection point information; wherein, the reflection point information includes at least the reflection point location and reflection intensity;

[0007] S2, Conversion and Preprocessing of Target Point Cloud: The reflection point information is converted into point cloud data, and the point cloud data is preprocessed.

[0008] S3, Extracting effective reflector coordinates based on distance gradient weights: By calculating the distance gradient between each reflection point and the lidar device, and using weights to weight the distance gradient, the effective reflector coordinates are extracted.

[0009] S4, Reflector Feature Recognition: Feature recognition is performed on the extracted reflector coordinates;

[0010] S5, Abnormal Reflective Feature Identification: Abnormal reflective features are identified by comparing the characteristics of reflective objects;

[0011] S6, Alarm and Record: The detected abnormal reflective features are processed as alarms, and then recorded and analyzed.

[0012] Optionally, the S2 transformation and preprocessing of the target point cloud specifically includes:

[0013] S21, extract the LiDAR scanning point information provided by the LiDAR driver and convert it into a point cloud;

[0014] S22, downsample the point cloud as needed;

[0015] S23, Traverse the point cloud according to requirements and exclude outliers;

[0016] S24, cluster the point cloud after excluding outliers, and extract the location information and index of each cluster.

[0017] S25. Based on the requirements, clusters are selected and excluded if the volume or number does not meet the requirements, and the target point cloud is determined.

[0018] Optionally, S3 extracts the effective reflector coordinates based on the distance gradient weights, specifically including:

[0019] The credibility of reflective object attributes is initially determined by using distance weights based on the real environment. A distance-light intensity evaluation index is established for the credibility of point cloud clustering that is recognized as valid after reflective object processing. The distance-light intensity evaluation index is calculated by the weighted distance of the cluster and the cluster light intensity to enable comparison of reflective objects at different distances under the same standard. The weighted distance is an exponential function of the distance between the target cluster coordinates and the coordinates of the center point of the lidar. The constant part of the exponential function is determined by environmental information and the main reflective object types in the scene, while the base part of the exponential function is determined by the set accuracy requirements and the resolution of the lidar at the set distance.

[0020] Optionally, S4 reflector feature recognition specifically includes:

[0021] S41, establish the contour information of each reflector by convex hull fitting;

[0022] S42, establish secondary clustering based on the contour information of each reflector;

[0023] S43, filter and classify secondary clustering according to requirements;

[0024] S44. Based on the clustering results, classify reflective objects according to requirements.

[0025] Optionally, S41 establishes the contour information of each reflector through convex hull fitting, specifically including: constructing convex hull fitting by traversing the points within the same cluster, and determining the cluster contour, the farthest heel pair, and the lower diameter of the pair through the fitting.

[0026] Optionally, S42 establishes a secondary cluster by expanding the search range and minimum number of clusters, so that clusters with a preset distance are merged into a new secondary cluster.

[0027] Optionally, S43 filters and classifies secondary clusters according to requirements, specifically including: performing convex hull fitting on the secondary clusters, comparing the convex hull information of the new clusters to obtain the subordinate relationship between the secondary clusters and the primary clusters; comparing the coordinate tables of the two clusters to determine the change in the centroid of the clusters caused by the secondary clusters; when the coordinate offset value brought about by cluster merging is lower than the threshold, the primary clusters that are far from the new centroids are classified as the abnormal reflective body clusters; where the distance is defined as Euclidean distance.

[0028] S44 compares the clustering results according to the requirements and classifies reflective bodies specifically, including: based on the secondary clustering classification results, the reflective body clusters obtained in S43 are divided into abnormal reflective body clusters and normal reflective body clusters.

[0029] Optionally, the S5 abnormal reflectivity feature identification specifically includes clustering the objects classified as abnormal reflective bodies, re-indexing their corresponding initial points, returning to the reflective body detection system step to extract the effective reflective body coordinates based on the distance gradient weights, and processing the points to which the abnormal reflective bodies belong; when the results of the new processing do not change, the attributes of the abnormal reflective bodies are determined, and the cause of their generation is confirmed by field testing; when the new processing causes the clustering results to change or the reflective bodies generate new classifications, the point cloud indexes in each iteration are manually operated and confirmed, and the attributes of the reflective bodies are confirmed by manual identification and field testing.

[0030] Optionally, the S6 alarm and logging specifically includes recording the secondary clustering table and analyzing the sudden changes in the number of clusters; when the number of clusters increases or decreases below the threshold, the system is considered to be operating normally; when the number of clusters changes and then returns to the original number or the change is large, the secondary clustering table is saved separately and an alarm is triggered in the program.

[0031] Secondly, a reflective object detection system based on lidar is provided, the system comprising:

[0032] An extraction module is used to acquire reflected signals in the environment through a lidar device and extract reflection point information; wherein, the reflection point information includes at least the reflection point location and reflection intensity;

[0033] The conversion module is used to convert the reflection point information into point cloud data and perform preprocessing operations on the point cloud data.

[0034] The extraction module is used to extract the effective coordinates of the reflector by calculating the distance gradient between each reflection point and the lidar device, and by weighting the distance gradient with weights.

[0035] The reflector feature recognition module is used to perform feature recognition on the extracted reflector coordinates;

[0036] An abnormal reflectivity feature recognition module is used to identify abnormal reflectivity features by comparing the features of reflective objects.

[0037] The alarm and recording module is used to process alarm information on detected abnormal reflective features, and to record and analyze such information.

[0038] Existing methods for locating reflective pillars cannot accurately determine the properties of the reflective objects, and the handling of coordinate errors of the reflective objects relies on subsequent positioning algorithms. This invention integrates the preprocessing and clustering processes into a single module, simplifying the algorithm structure; it incorporates screening and processing for anomalies, improving the algorithm's accuracy; it adds an analysis and anomaly alarm system, enhancing the algorithm's applicability in complex environments; and it reduces operational difficulty and improves the deployment efficiency of the positioning algorithm. Attached Figure Description

[0039] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0040] Figure 1 A flowchart of a reflective object detection method based on lidar provided in this application embodiment. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0042] In the description of this invention, the terms “comprising,” “having,” and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are expressly listed, but may also include other steps or units that are not expressly listed but are inherent to these processes, methods, products, or apparatuses, or steps or units added based on further optimizations of the inventive concept.

[0043] The purpose of this application is to determine the attributes of reflective objects by establishing their feature information using lidar, including category, coordinates, range of associated points, and positioning reliability. This provides additional information to improve positioning accuracy and reduce the difficulty of field testing the positioning algorithm. For details, please refer to... Figure 1 The diagram illustrates a flowchart of a reflective object detection method based on lidar provided in an embodiment of this application. The method may include the following steps:

[0044] S1, Extracting Reflection Point Information from LiDAR Signals: Obtaining reflection signals from the environment using LiDAR equipment and extracting reflection point information.

[0045] This step specifically implements the acquisition of lidar signals. The reflection point information includes at least the reflection point location and reflection intensity.

[0046] S2, Transformation and Preprocessing of Target Point Cloud: Converts reflection point information into point cloud data and performs preprocessing operations on the point cloud data.

[0047] This step specifically implements the preprocessing procedure, specifically:

[0048] The transformation and preprocessing of target point clouds specifically includes:

[0049] S21, extract the LiDAR scanning point information provided by the LiDAR driver and convert it into a point cloud;

[0050] S22, downsample the point cloud as needed;

[0051] S23, Traverse the point cloud according to requirements and exclude outliers;

[0052] S24, cluster the point cloud after excluding outliers, and extract the location information and index of each cluster.

[0053] S25. Based on the requirements, clusters are selected and excluded if the volume or number does not meet the requirements, and the target point cloud is determined.

[0054] In this embodiment of the application, requirement information can be added in advance. The requirement information includes at least the overall downsampling of the point cloud, the index for excluding outliers, and the information for filtering clustering conditions.

[0055] S3, Extracting effective reflector coordinates based on distance gradient weights: By calculating the distance gradient between each reflection point and the lidar device, and using weights to weight the distance gradient, the effective reflector coordinates are extracted.

[0056] In extracting effective reflector coordinates based on distance gradient weights: The reliability of reflector attributes is initially determined using distance weights based on the real-world environment. An evaluation index, called the distance-intensity evaluation index, is established to assess the reliability of point cloud clusters that are considered effective after reflector processing. This index is calculated from the weighted distance of the clusters and the cluster intensity, allowing for comparison of reflectors at different distances using the same standard. The weighted distance is an exponential function of the distance between the target cluster coordinates and the coordinates of the LiDAR center point. The constant part of this exponential function is determined by environmental information and the main reflector types in the scene, while the base part is determined by the accuracy requirements of this algorithm and the radar's performance at medium to long range resolution. Depending on the application scenario, when the downstream algorithm needs to pay extra attention to targets within a specific medium-range, a distance-based gradient weight needs to be added to adjust the effective cluster range. This weight will be reflected in the constant part of the exponential function. The cluster intensity is the intensity value returned by the LiDAR reflection point. In this step, clusters whose distance-intensity evaluation index does not reach the threshold are excluded; the remaining clusters are considered effective reflectors. The threshold is determined by the scenario and accuracy requirements.

[0057] S4, Reflector Feature Recognition: Perform feature recognition on the extracted reflector coordinates.

[0058] The specific steps for identifying reflective features in this process include:

[0059] S41, establish the contour information of each reflector by convex hull fitting;

[0060] S42, establish secondary clustering based on the contour information of each reflector;

[0061] S43, filter and classify secondary clustering according to requirements;

[0062] S44. Based on the clustering results, classify reflective objects according to requirements.

[0063] Specifically, in optional embodiments of this application, establishing the contour information of each reflector through convex hull fitting specifically includes: constructing convex hull fitting by traversing the points within the same cluster, and determining the cluster contour, the farthest pair of heel points, and the lower diameter of the pair through the fitting.

[0064] In an optional embodiment of this application, establishing a secondary cluster specifically includes: expanding the search range and minimum number of clusters so that clusters with a preset distance are merged into a new secondary cluster.

[0065] In optional embodiments of this application, the screening and classification of secondary clusters according to requirements specifically includes: performing convex hull fitting on the secondary clusters, comparing the convex hull information of the new clusters to obtain the subordinate relationship between the secondary clusters and the primary clusters; comparing the coordinate tables of the two clusters to determine the change in the centroid of the clusters caused by the secondary clusters; when the coordinate offset value brought about by cluster merging is lower than the threshold, the primary clusters that are far from the new centroid are classified as the abnormal reflective body clusters; wherein, the distance is defined as Euclidean distance;

[0066] In optional embodiments of this application, classifying reflective bodies specifically includes: based on the secondary clustering classification results, dividing the reflective body clusters obtained in S43 into abnormal reflective body clusters and normal reflective body clusters.

[0067] S5, Abnormal Reflective Feature Identification: Abnormal reflective features are identified by comparing the characteristics of reflective objects.

[0068] In the abnormal reflectivity feature identification step: for clusters classified as abnormal reflective objects, the initial points are re-indexed, and the system returns to the reflective object detection step to extract valid reflective object coordinates based on distance gradient weights, processing the points belonging to the abnormal reflective objects. If the new processing results do not change, the abnormal reflective object attributes are determined, and the cause of its generation is confirmed through field testing. If the new processing causes changes in the clustering results or a new classification of reflective objects, the point cloud indexes in each iteration are manually checked, and the reflective object attributes are confirmed through manual identification and field testing.

[0069] S6, Alarm and Record: The detected abnormal reflective features are processed as alarms, and then recorded and analyzed.

[0070] The specific process of alarming and recording in this step includes recording the secondary clustering table and analyzing sudden changes in the number of clusters. When the number of clusters increases or decreases below a threshold, the system is considered to be operating normally; when the number of clusters changes drastically and then returns to its original value or the change is significant, the secondary clustering table is saved separately and an alarm is triggered in the program.

[0071] The above steps will be further explained below with reference to an embodiment proposed in this application:

[0072] A lidar data acquisition platform can be formed using lidar, industrial control computers, and other devices. On this platform, the lidar scanning frequency and position can be set, and lidar scanning data can be collected.

[0073] The collected lidar reflection points are converted into lidar point clouds that can be further manipulated, where each point in the point cloud should contain coordinate and light intensity information.

[0074] The point cloud is downsampled overall. This step can be skipped depending on the radar performance and algorithm requirements.

[0075] Traverse the point cloud and exclude a small number of outliers and other anomalies.

[0076] For the current point cloud clustering, establish information for each cluster, including the cluster location, the number of points contained in the cluster, and the index of the points contained in the cluster.

[0077] Clusters are screened, excluding those whose volume or number does not meet the requirements.

[0078] A distance-intensity evaluation index is calculated from the weighted distance and light intensity of clusters, and additional weighting rules are added based on the downstream algorithm's focus on specific ranges. Clusters whose distance-intensity evaluation index does not reach the scene threshold are excluded, and the remaining clusters can be identified as valid reflectors.

[0079] Contour hull fitting is used to establish the contour information of each reflector, and the points that form the contour and the geometric information of the contour are recorded.

[0080] A second clustering is performed on the aforementioned clustering results, which requires additional parameter adjustments.

[0081] Convex hull fitting is performed on the secondary clusters, and the morphological information is compared to obtain the hierarchical relationship between the secondary and primary clusters. The cluster merging process reflected by the hierarchical relationship is determined. The dominant category in the analysis results is considered an effective reflective body, while the rest are considered an anomalous reflective bodies.

[0082] The steps following the distance gradient weighting are repeated for anomalous reflectors. If the new process cannot eliminate the reflector, it is identified as an anomalous reflector. When the definition of the reflector changes due to the new process, the cause of the anomalous reflector's generation and the algorithm process are manually verified.

[0083] Record the secondary clustering table, record and alert for cases where mutations are followed by recovery, stop the current localization algorithm and test to eliminate the problem.

[0084] This application also provides a reflective object detection system based on lidar. The system includes:

[0085] An extraction module is used to acquire reflected signals in the environment through a lidar device and extract reflection point information; wherein, the reflection point information includes at least the reflection point location and reflection intensity;

[0086] The conversion module is used to convert the reflection point information into point cloud data and perform preprocessing operations on the point cloud data.

[0087] The extraction module is used to extract the effective coordinates of the reflector by calculating the distance gradient between each reflection point and the lidar device, and by weighting the distance gradient with weights.

[0088] The reflector feature recognition module is used to perform feature recognition on the extracted reflector coordinates;

[0089] An abnormal reflectivity feature recognition module is used to identify abnormal reflectivity features by comparing the features of reflective objects.

[0090] The alarm and recording module is used to process alarm information on detected abnormal reflective features, and to record and analyze such information.

[0091] For specific limitations on lidar-based reflective object detection systems, please refer to the limitations on lidar-based reflective object detection methods mentioned above, which will not be repeated here.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for detecting reflective objects based on lidar, characterized in that, The method includes: S1, Extracting reflection point information from lidar signals: Obtaining reflection signals in the environment through lidar equipment and extracting reflection point information; wherein, the reflection point information includes at least the reflection point location and reflection intensity; S2, Conversion and Preprocessing of Target Point Cloud: The reflection point information is converted into point cloud data, and the point cloud data is preprocessed. S3, Extracting effective reflector coordinates based on distance gradient weights: By calculating the distance gradient between each reflection point and the lidar device, and using weights to weight the distance gradient, the effective reflector coordinates are extracted. S4, Reflector Feature Recognition: Feature recognition is performed on the extracted reflector coordinates; S5, Abnormal Reflective Feature Identification: Abnormal reflective features are identified by comparing the characteristics of reflective objects; S6, Alarm and Record: The detected abnormal reflective features are processed as alarms, and then recorded and analyzed; S2 transformation and preprocessing of target point clouds specifically includes: S21, extract the LiDAR scanning point information provided by the LiDAR driver and convert it into a point cloud; S22, downsample the point cloud as needed; S23, Traverse the point cloud according to requirements and exclude outliers; S24, cluster the point cloud after excluding outliers, and extract the location information and index of each cluster. S25, based on the requirements, select clusters, exclude clusters that do not meet the volume or quantity requirements, and determine the target point cloud; S3 extracts the effective reflector coordinates based on distance gradient weights, specifically including: The credibility of reflective object attributes is initially determined by using distance weights based on the real environment. A distance-light intensity evaluation index is established for the credibility of point cloud clustering that is recognized as valid after reflective object processing. The distance-light intensity evaluation index is calculated from the weighted distance of the cluster and the cluster light intensity to enable comparison of reflective objects at different distances under the same standard. The weighted distance is an exponential function of the distance between the target cluster coordinates and the coordinates of the center point of the lidar. The constant part of the exponential function is determined by environmental information and the main reflective object types in the scene, while the base part of the exponential function is determined by the set accuracy requirements and the resolution of the lidar at the set distance. S4 reflector feature identification specifically includes: S41, establish the contour information of each reflector by convex hull fitting; S42, establish secondary clustering based on the contour information of each reflector; S43, filter and classify secondary clustering according to requirements; S44, classify reflective objects by comparing clustering results according to requirements; S43, based on requirements, filters and classifies secondary clusters, specifically including: fitting the convex hull of the secondary clusters; comparing the convex hull information of the new clusters to obtain the subordinate relationship between the secondary and primary clusters; comparing the coordinate tables of the two clusters to determine the change in the centroid of the clusters caused by the secondary clusters; when the coordinate offset value caused by cluster merging is lower than the threshold, the primary clusters that are far from the new centroid are classified as the abnormal reflective body cluster; where the distance is defined as Euclidean distance. S44 compares the clustering results according to the requirements and classifies reflective bodies specifically, including: based on the secondary clustering classification results, the reflective body clusters obtained in S43 are divided into abnormal reflective body clusters and normal reflective body clusters.

2. The method according to claim 1, characterized in that, S41 establishes the contour information of each reflector by convex hull fitting, specifically by: traversing the points within the same cluster to establish convex hull fitting, and using this fitting to determine the contour of the cluster, the farthest heel pair, and the lower diameter of the pair.

3. The method according to claim 1, characterized in that, S42 establishes a secondary cluster by expanding the search range and minimum number of clusters, so that clusters with a preset distance are merged into a new secondary cluster.

4. The method according to claim 1, characterized in that, The S5 abnormal reflectivity feature identification specifically includes clustering objects classified as abnormal reflective bodies, re-indexing their corresponding initial points, returning to the reflective body detection system steps to extract valid reflective body coordinates based on distance gradient weights, and processing the points to which the abnormal reflective bodies belong. When the results of the new processing do not change, the attributes of the abnormal reflective bodies are determined, and the cause of their generation is confirmed through field testing. When the new processing causes changes in the clustering results or the reflective bodies generate new classifications, the point cloud indexes in each iteration are manually operated and confirmed, and the attributes of the reflective bodies are confirmed through manual identification and field testing.

5. The method according to claim 1, characterized in that, S6 alarms and logging specifically include recording the secondary clustering table and analyzing sudden changes in the number of clusters; when the number of clusters increases or decreases below the threshold, the system is considered to be operating normally; when the number of clusters changes and then returns to the original number or the change is large, the secondary clustering table is saved separately and an alarm is triggered in the program.

6. A reflective object detection system based on lidar, characterized in that, The system includes: An extraction module is used to acquire reflected signals in the environment through a lidar device and extract reflection point information; wherein, the reflection point information includes at least the reflection point location and reflection intensity; The conversion module is used to convert the reflection point information into point cloud data and perform preprocessing operations on the point cloud data. The extraction module is used to extract the effective coordinates of the reflector by calculating the distance gradient between each reflection point and the lidar device, and by weighting the distance gradient with weights. The reflector feature recognition module is used to perform feature recognition on the extracted reflector coordinates; An abnormal reflectivity feature recognition module is used to identify abnormal reflectivity features by comparing the features of reflective objects. The alarm and recording module is used to process the detected abnormal reflective features with alarms, and to record and analyze them. The transformation and preprocessing of target point clouds specifically includes: Extract the LiDAR scan point information provided by the LiDAR driver and convert it into a point cloud; Downsampling of the point cloud as needed; Traverse the point cloud according to requirements and exclude outliers; Cluster the point cloud after excluding outliers, and extract the location information and index of each cluster. Clusters are selected based on requirements, and clusters that do not meet the volume or quantity requirements are excluded to determine the target point cloud. Extracting the effective reflector coordinates based on distance gradient weights specifically includes: The credibility of reflective object attributes is initially determined by using distance weights based on the real environment. A distance-light intensity evaluation index is established for the credibility of point cloud clustering that is recognized as valid after reflective object processing. The distance-light intensity evaluation index is calculated from the weighted distance of the cluster and the cluster light intensity to enable comparison of reflective objects at different distances under the same standard. The weighted distance is an exponential function of the distance between the target cluster coordinates and the coordinates of the center point of the lidar. The constant part of the exponential function is determined by environmental information and the main reflective object types in the scene, while the base part of the exponential function is determined by the set accuracy requirements and the resolution of the lidar at the set distance. Reflective feature recognition specifically includes: Contour hull fitting is used to establish the contour information of each reflector; A secondary clustering was established based on the outline information of each reflector; Filter and classify based on requirements, then perform secondary clustering; Based on the clustering results compared to the requirements, reflective objects are classified. The specific steps for filtering and classifying secondary clusters based on requirements include: fitting the convex hull of the secondary clusters; comparing the convex hull information of the new clusters to determine the hierarchical relationship between the secondary and primary clusters; comparing the coordinate tables of the two clusters to determine the change in the centroid of the clusters caused by the secondary clusters; when the coordinate offset value caused by cluster merging is lower than the threshold, the primary clusters that are far from the new centroid are classified as the anomalous reflective clusters; where the distance is defined as Euclidean distance. Based on the clustering results compared to the requirements, the classification of reflective bodies specifically includes: based on the secondary clustering classification results, the obtained reflective bodies are clustered into abnormal reflective body clusters and normal reflective body clusters.

Images