Search

CN-121982660-A - Multi-frame lane central line fusion method for highway side monitoring

CN121982660ACN 121982660 ACN121982660 ACN 121982660ACN-121982660-A

Abstract

The invention provides a multi-frame lane central line fusion method for highway side monitoring, which relates to the technical field of data processing, and comprises the following steps: and (3) inputting all lane center line path point sets detected by the current frame to carry out resampling processing, traversing each path to carry out overlap ratio calculation and fusion path matching judgment, dismantling the current path, if the fusion path matching judgment result is yes, adding a plurality of local fragments into a fragment pool of the fusion path to carry out life cycle management and fragment deduplication, carrying out connection relation construction and path search scoring to obtain a plurality of groups of continuous fragment sequence scoring, carrying out fragment sequence selection and path connection reconstruction, and outputting a lane center line fusion path. The technical problems of poor accuracy and stability of a lane center line fusion path in the prior art are solved. The technical effects of improving the accuracy and stability of the lane central line fusion path under the highway side monitoring are achieved.

Inventors

  • YAN JUN
  • ZHANG YU

Assignees

  • 超级视线科技有限公司

Dates

Publication Date
20260505
Application Date
20260109

Claims (9)

  1. 1. A multi-frame lane centerline fusion method for highway side monitoring, the method comprising: inputting all lane center line path point sets detected by the current frame, resampling the lane center line path point sets, traversing each path, and carrying out overlap ratio calculation and fusion path matching judgment to obtain a fusion path matching judgment result; The current path is disassembled into a plurality of local fragments and a plurality of fragment geometric attributes are recorded, if the fusion path is matched and judged, the plurality of local fragments are added into a fragment pool of the fusion path to carry out life cycle management and fragment deduplication, and a path updating fragment pool is obtained; Performing connection relation construction and path search scoring for each fragment in the path updating fragment pool according to the geometric attributes of the fragments to obtain a plurality of groups of continuous fragment sequence scores; and selecting a segment sequence, reconstructing path connection based on the scores of the multiple groups of continuous segment sequences, and outputting a lane center line fusion path.
  2. 2. A multi-frame lane centerline fusion method for highway side monitoring as claimed in claim 1, the method is characterized by comprising the steps of traversing each path to perform overlap ratio calculation and fusion path matching judgment, and comprises the following steps: constructing an overlap ratio calculation formula, wherein the overlap ratio calculation formula specifically comprises the following steps: ; ; ; Wherein the new detection path is The fusion path traversed currently is , And Respectively representing a set of path points And A line segment formed by every two points of the line segment, Is the midpoint of the line segment, In order to approach the threshold value, Is an indication function; Traversing each path according to the overlap ratio calculation formula to perform overlap ratio calculation, and obtaining an overlap ratio calculation result of all paths; and comparing the calculation result of the overlapping proportion of all paths with a preset overlapping proportion threshold value to be used as a fusion path matching judgment result.
  3. 3. The multi-frame lane centerline fusion method for highway side monitoring as claimed in claim 1, wherein the breaking down the current path into a plurality of local segments and recording a plurality of segment geometric attributes comprises: starting the starting point of the current path, sequentially constructing path segments according to a point sequence, regarding the segments as the extension of the same segment when the direction change between continuous points does not exceed a preset threshold value and the number of points in the segments does not exceed a maximum limit, ending the current segment and restarting segment construction from the ending position to obtain a plurality of local segments if the direction mutation exceeds the preset threshold value or the number of points in the segments reaches the maximum limit; And carrying out attribute recording on each generated segment in the plurality of local segments to obtain a plurality of segment geometric attributes, wherein the plurality of segment geometric attributes comprise a starting point, a direction vector and a segment length.
  4. 4. The method of multi-frame lane centerline fusion for highway side monitoring as claimed in claim 1, wherein obtaining a pool of path update segments comprises: Adding the plurality of local fragments into a fragment pool of a fusion path to perform life cycle management to obtain a path dynamic fragment pool, wherein the life cycle management comprises frame number recording of frames to which the fragments belong and fragment pool cleaning operation; introducing a hash function based on quantization coordinates to carry out space quantization on the plurality of local fragments in the path dynamic fragment pool to obtain a plurality of path Duan Haxi values; And carrying out hash bucket mapping and equivalence judgment based on the paths Duan Haxi values, and carrying out fragment deduplication on the path dynamic fragment pool to obtain a path updating fragment pool.
  5. 5. The method of multi-frame lane centerline fusion for highway side monitoring as claimed in claim 4, wherein obtaining a plurality of path Duan Haxi values comprises: introducing a hash function based on quantized coordinates to spatially quantize start and end coordinates of the plurality of local segments in the path dynamic segment pool, and mapping continuous values to a fixed grid: Where r is the quantization resolution, 、 Respectively obtaining quantized coordinates of a plurality of path segments, namely a starting point and an ending point of the path segment S: ; and quantizing the coordinates by using the plurality of path segments, and combining to obtain a plurality of path Duan Haxi values: , wherein, In the case of an integer hash of the data, Is bitwise exclusive or.
  6. 6. The method of claim 1, wherein obtaining a plurality of sets of sequential segment sequence scores comprises: Calculating forward connection and backward connection for each segment in the path update segment pool to form a forward bidirectional adjacency list and a backward bidirectional adjacency list; Constructing a fragment scoring formula: , wherein, For the length of the segment(s), For the direction angle of the candidate segment direction to the current constructed path, And Each weight coefficient; and according to the multiple fragment geometric attributes, based on the forward bidirectional adjacency list and the backward bidirectional adjacency list and the fragment scoring formula, taking any segment in the path updating fragment pool as a starting point, and simultaneously carrying out forward and backward beam searching and scoring accumulation to obtain multiple groups of continuous fragment sequence scores.
  7. 7. A multi-frame lane centerline fusion method for highway side monitoring as claimed in claim 2, wherein said method further comprises: if the fusion path matching judgment result is negative, a new detection path is detected Matching with each path to be judged to obtain a path matching result to be judged; If the result of the to-be-judged path matching is yes, adding the plurality of local fragments into a fragment pool of the to-be-judged path, adding 1 to the number of times of the to-be-judged path matching, and reconstructing the to-be-judged path from the fragment pool.
  8. 8. The method for multi-frame lane centerline fusion for highway side monitoring according to claim 7, wherein after reconstructing the pending path from the segment pool, comprising: And judging whether the matching times of the paths to be judged are greater than a preset threshold value, if so, converting the paths to be judged into fusion paths, outputting lane center line fusion paths, and if not, directly outputting the lane center line fusion paths.
  9. 9. The multi-frame lane centerline fusion method for highway side monitoring as claimed in claim 7, wherein said method further comprises: if the matching results of the paths to be judged are not matched, recording the new detection path The method comprises the steps of taking a new path to be judged, and disassembling the new path to be judged into a plurality of continuous fragments; And adding the continuous fragments into a newly created fragment pool of the new to-be-judged path, and outputting a lane central line fusion path.

Description

Multi-frame lane central line fusion method for highway side monitoring Technical Field The invention relates to the technical field of data processing, in particular to a multi-frame lane central line fusion method for highway side monitoring. Background In a highway roadside monitoring system, a high-level camera can continuously capture traffic flow information of multiple lanes, and high-precision lane center line data is generally required to support tasks such as traffic operation monitoring, lane occupation identification, flow statistics and anomaly detection. Due to the complex factors of deviation of camera view angle, long-distance compression, bad weather, vehicle shielding and the like, the lane center line extracted from a single frame image often accompanies jitter, short-term loss or local zigzag error. Therefore, the prior art is generally based on single-frame lane line detection, and the continuity and stability of the lane center line are improved by extracting lane center line candidate paths frame by frame and then fusing multi-frame data by a simple time filtering or morphological matching method. However, in the road side scene, due to the characteristics of large traffic flow, frequent shielding condition, obvious degradation of the lane lines at a long distance and the like, the shape change of the path detection result is more complex, especially the local shape change of the lane, such as overlapping, opposite directions or small section missing, and the like, so that the difficulty of multi-frame fusion is obviously increased, and the accuracy and stability of the fusion path are affected. The prior art has the technical problems that the local shape of the lane is changeable due to complex factors under the expressway side scene, the multi-frame fusion difficulty is high, and the accuracy and the stability of the lane center line fusion path are affected. Disclosure of Invention The application aims to provide a multi-frame lane central line fusion method for highway side monitoring, which is used for solving the technical problems that the local forms of lanes are changeable, the multi-frame fusion difficulty is high, and the accuracy and the stability of lane central line fusion paths are influenced because of complex factors under a highway side scene are difficult to effectively process in the prior art. In view of the above problems, the application provides a multi-frame lane center line fusion method for highway side monitoring, which comprises the steps of inputting all lane center line path point sets detected by a current frame, resampling all lane center line path point sets, traversing each path, performing overlap ratio calculation and fusion path matching judgment to obtain a fusion path matching judgment result, disassembling the current path into a plurality of local fragments, recording a plurality of fragment geometric attributes, if the fusion path matching judgment result is yes, adding the plurality of local fragments into a fragment pool of a fusion path for life cycle management and fragment deduplication to obtain a path update fragment pool, constructing a connection relation and path search scores for each fragment in the path update fragment pool according to the plurality of fragment geometric attributes, obtaining a plurality of groups of continuous fragment sequence scores, selecting a fragment sequence, reconstructing path connection based on the plurality of groups of continuous fragment sequence scores, and outputting a lane center line fusion path. Optionally, an overlap ratio calculation formula is constructed, where the overlap ratio calculation formula specifically includes:;; ; Wherein the new detection path is The fusion path traversed currently is,AndRespectively representing a set of path pointsAndA line segment formed by every two points of the line segment,Is the midpoint of the line segment,In order to approach the threshold value,The method comprises the steps of obtaining a path overlapping proportion calculation result, obtaining an indication function, traversing each path according to the overlapping proportion calculation formula to obtain an overlapping proportion calculation result of all paths, and using a comparison result of the overlapping proportion calculation result of all paths and a preset overlapping proportion threshold value as a fusion path matching judgment result. Optionally, starting the starting point of the current path, sequentially constructing path segments according to a point sequence, when the direction change between continuous points does not exceed a preset threshold value and the number of points in the segments does not exceed a maximum limit, regarding the segment as the extension of the same segment, ending the current segment and restarting segment construction from the ending position to obtain a plurality of local segments if the direction change exceeds the preset threshold value or the numbe