CN-121982887-A - Traffic perception system and traffic perception method based on multi-source information fusion

CN121982887ACN 121982887 ACN121982887 ACN 121982887ACN-121982887-A

Abstract

The application relates to the technical field of traffic information fusion, and discloses a traffic perception system and a traffic perception method based on multi-source information fusion, wherein the traffic perception system comprises a plurality of radar vehicle detectors; the system comprises a monitoring upright rod, a plurality of MEC edge computing units, a radar vehicle detector, one or more MEC edge computing units and a plurality of MEC edge computing units, wherein each MEC edge computing unit is in signal connection with a millimeter wave radar and a video camera corresponding to the monitoring upright rod and corresponds to a plurality of radar vehicle detectors, the radar vehicle detectors are connected with the one or more MEC edge computing units, and the MEC edge computing units conduct multi-source information fusion on information of all the millimeter wave radar, the video camera and the radar vehicle detectors to conduct full track monitoring of a key node section. According to the traffic perception system and the traffic perception method, multi-view traffic information perception is achieved, the perception capability of a key road section is improved, and the technical problems that accurate perception cannot be achieved in a scene of a front cart which is shielded and a back cart, errors are large in a scene with a far distance or false targets appear are solved.

Inventors

ZHENG YIFAN
ZHAO QIANQIAN
GU QIANWEI
HU LINGYU
LU DONGLIN
WU YUEWEN
FENG NING
WANG JUAN
ZHENG TENGKUN
LEI MING

Assignees

中铁大桥勘测设计院集团有限公司

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (10)

1. A traffic perception system based on multi-source information fusion comprises millimeter wave radar and video camera positioned in monitoring upright posts, each monitoring upright post corresponds to two upright post perception areas, and is characterized in that, The traffic perception system comprises: the radar vehicle detectors are arranged on the road side anti-collision guardrail at intervals, and at least one radar vehicle detector is arranged in each upright rod sensing area and in a handshake area between every two adjacent upright rod sensing areas; The MEC edge computing units are arranged on the monitoring vertical rods one by one, and each MEC edge computing unit is connected with a millimeter wave radar and a video camera corresponding to the monitoring vertical rod, a radar car detector in two vertical rod sensing areas corresponding to the monitoring vertical rod, a radar car detector in a handshake area between the two vertical rod sensing areas corresponding to the monitoring vertical rod, and a radar car detector in a handshake area between the two vertical rod sensing areas corresponding to the monitoring vertical rod and the front monitoring vertical rod and the rear monitoring vertical rod; and the MEC edge computing units fuse the information of all millimeter wave radars, video cameras and radar car detectors with multi-source information to perform full track monitoring of the key node road sections.
2. The traffic perception system based on multi-source information fusion as set forth in claim 1, wherein the radar vehicle detector is a laser radar vehicle detector or a microwave radar vehicle detector; the handshake area comprises an own sensing blank area between two vertical rod sensing areas under each monitoring vertical rod, an interval sensing blank area between two sensing areas of two adjacent monitoring vertical rods and a weak sensing overlapping area of two sensing areas of two adjacent monitoring vertical rods; The radar vehicle detectors with the space perception blank areas or the weak perception overlapping areas are respectively connected with MEC edge computing units of the front and rear monitoring upright poles in a signal mode.
3. The traffic perception system based on multi-source information fusion as set forth in claim 1, wherein the traffic perception system comprises a cloud server, and the cloud server is configured to receive the multi-source information fusion traffic perception results transmitted by all MEC edge computing units and to aggregate the traffic perception results of the entire key node section.
4. A traffic perception method based on the traffic perception system as claimed in claim 1, characterized in that a bayonet video camera and a laser radar are arranged at the entrance of the key node section and in the 1 st vertical rod perception area, and the perception method comprises the following steps: S1, a vehicle enters a1 st upright post sensing area, a MEC edge computing unit fuses video data of a bayonet video camera and radar data of a laser radar, multisource information is fused with upright post video and radar data of the 1 st upright post sensing area, and vehicle information data of a current area is acquired after fusion, wherein the vehicle information data comprises a vehicle ID, and optimized fusion sensing target position coordinates and track data of a target vehicle; S2, entering a later handshake area, and acquiring vehicle information data of a current area by fusing vehicle information data detected by a vehicle detector sensing area of a multi-source information fusion radar vehicle detector and fused vehicle information data of a previous upright rod sensing area; S3, entering a rear upright sensing area, wherein the multisource information is fused with vehicle information data detected by a vehicle detector sensing area of a radar vehicle detector, fused vehicle information data of a front upright sensing area, fused vehicle information data of a front handshake area, upright video and radar data of an own upright sensing area, and acquiring vehicle information data of a current area after fusion; And S4, repeating the steps S2 and S3 until the vehicle exits the key node road section.
5. The method according to claim 4, wherein step S1 comprises the steps of performing multi-source information fusion on video data of the bayonet video camera and radar data of a laser radar at an entrance of the key node section and at an entrance of a1 st vertical rod sensing area by an MEC edge computing unit to obtain feature initialized vehicle information data; and optimizing the vehicle information data of other areas in the 1 st vertical pole sensing area by taking the vehicle information data initialized by the features as a reference and matching with the vertical pole video and radar data of the current vertical pole sensing area.
6. The method according to claim 4, wherein the handshake area is an autonomous sensing empty area between two upright sensing areas directly under the monitoring upright, a sensor sensing area of a radar sensor in the autonomous sensing empty area is located in the middle of the autonomous sensing empty area in the front-rear direction, and the front side and the rear side have two sensing blind areas, and the step S2 comprises: The method comprises the steps that a vehicle enters an own perception blank area, a MEC edge computing unit corresponding to the own perception blank area performs feature initialization on vehicle information data fused by a previous upright rod perception area of the MEC edge computing unit, the MEC edge computing unit is used as starting data of a first perception blind area, a vehicle state prediction model based on microscopic traffic simulation takes over a continuous vehicle track tracking task of the blind area, and predicted vehicle information data of the first perception blind area is obtained; The vehicle enters a sensing area of the vehicle detector, the corresponding MEC edge computing unit fuses detection information of the sensing area of the vehicle detector and predicted vehicle information data of a first sensing blind area by utilizing a multisource information fusion sensing algorithm, and continuously tracks and acquires the predicted vehicle information data of a second sensing blind area.
7. The method of claim 4, wherein the handshake area is a space sensing blank area between two adjacent sensing areas of two adjacent monitoring uprights, the sensing area of the radar car detector for the space sensing blank area is located in the middle of the front-back direction, and the front side and the back side have two sensing blind areas, and the step S2 comprises: the method comprises the steps that a vehicle enters an interval perception blank area, a MEC edge computing unit corresponding to the interval perception blank area, and feature initialization is carried out on vehicle information data fused by a previous upright rod perception area, wherein the vehicle information data is used as initial data of a first perception blind area, a vehicle state prediction model based on microscopic traffic simulation takes over a continuous vehicle track tracking task of the blind area, and predicted vehicle information data of the first perception blind area is obtained; The vehicle enters a sensing area of the vehicle detector, the target track information is corrected by using the detection information of the sensing area of the vehicle detector, the detection information of the sensing area of the vehicle detector and the predicted vehicle information data of the first sensing blind area are fused by using a multi-source information fusion sensing algorithm, and the predicted vehicle information data of the second sensing blind area is continuously tracked and obtained.
8. A sensing method of the sensing system of claim 4, wherein: the radar vehicle detector comprises a radar vehicle detector, a radar vehicle detector and a radar vehicle detector, wherein the radar vehicle detector is arranged in the radar vehicle detector; The step S2 includes: The vehicle enters a weak perception overlapping area, a MEC edge computing unit corresponding to a previous upright perceiving area acquires detection data of a radar vehicle detector of the weak perception overlapping area, acquires high-precision measurement information through fusion of upright video and radar data corresponding to the previous upright perceiving area and a fusion perceiving algorithm, positions a target vehicle driving away from the previous upright perceiving area and transmits the target vehicle to a MEC edge computing unit of a subsequent upright perceiving area through a specific protocol; And the MEC edge computing unit of the next vertical rod sensing area performs feature initialization on the received target information, and utilizes a fusion sensing algorithm to fuse vertical rod video and radar data corresponding to the next vertical rod sensing area to acquire vehicle information data of the next vertical rod sensing area.
9. The sensing method of the sensing system according to claim 4, further comprising, in step S3: At the entrance in the upright post sensing area, the MEC edge computing unit of the current upright post sensing area fuses the vehicle information data of the previous handshake area and upright post video and radar data of the current upright post sensing area to perform feature initialization; In the upright sensing area and in front of the sensing area of the vehicle detector, continuously tracking the target based on the initialized vehicle information data; In the vertical rod sensing area, in and after the vehicle detector sensing area, the MEC edge computing unit fuses vehicle information data detected by the radar vehicle detector and vertical rod video and radar data of the current vertical rod sensing area, and optimizes the vehicle information data in and after the vehicle detector sensing area.
10. The sensing method of the sensing system according to claim 9, wherein a local three-dimensional coordinate system is established by taking a base of each monitoring upright as an origin, and local coordinate positions of all radar detectors, millimeter wave radar and video cameras are obtained through calibration in advance; All the positions of the target vehicles acquired by the radar vehicle detectors, the millimeter wave radar and the video camera are converted into local coordinate positions; When the vehicle information data of the previous MEC edge computing unit is transmitted to the next MEC edge computing unit, local coordinate conversion is automatically performed.

Description

Traffic perception system and traffic perception method based on multi-source information fusion Technical Field The invention relates to the technical field of traffic information fusion, in particular to a traffic perception system and a traffic perception method based on multi-source information fusion. Background Along with the development of traffic informatization and vehicle-road coordination, the real-time traffic perception and monitoring of all road sections becomes an essential block of an intelligent traffic system, and especially in a plurality of huge bridges, tunnels and partial group fog and severe weather multiple road sections crossing the river and the sea, the global perception of traffic can provide basic data support for traffic guidance or management and control. In the related art, the scheme of the traditional perception system is mainly based on a plurality of monitoring vertical rods, each monitoring vertical rod is in a shape of a door or an inverted L, each monitoring vertical rod is inserted into the ground through a vertical rod, a millimeter wave radar and a video camera are fixed through a cross rod, and the plurality of monitoring vertical rods are arranged at intervals. However, there are significant drawbacks to the conventional perception system scheme: On some important freight passages, the proportion of container trucks and trailers on bridges and tunnel sections is high, traffic accidents are easy to occur, the height of large vehicles is usually higher than that of small vehicles, physical shielding exists, and a certain angle exists on many road surfaces crossing the river and the sea, so that the shielding effect is more remarkable. In a scene of a front cart shielding a rear cart (namely, the front cart is arranged on the front side or the inclined front side of the rear cart), a video camera always exists in a traffic sensing system, the front cart (such as a freight container vehicle) shields the rear cart, so that the detection is difficult, in the scene of a millimeter wave radar, the millimeter wave radar encounters a metal object with a large and regular appearance, which is equivalent to the situation of encountering a strong reflector, and the millimeter wave radar can detect flickering or vanishing jump to the cart beside the cart and the remote cart, so that the video camera and the millimeter wave radar cannot accurately sense in the scene of the front cart shielding the rear cart. Further, in a far-distance scene, the perceived distance of a video camera is extremely limited, the video camera is difficult to estimate at a slightly long distance and has large error, and in the far-distance scene, a millimeter wave radar is easy to generate a false target due to serious multipath problems, specifically, multipath refers to the propagation phenomenon that a radar transmitting signal reaches a measured target from a transmitting antenna through a plurality of paths, a target echo reaches a receiving antenna through a plurality of paths, and more metal structures exist in a river-crossing and sea-crossing oversized bridge and tunnel, and at the moment, the multipath effect in the radar detection process is increased to generate the false detection target. Disclosure of Invention The application provides a traffic perception system and a traffic perception method based on multi-source information fusion, which are characterized in that a plurality of radar car detectors are newly added at a road side, millimeter wave radar, video cameras and radar car detector data are fused to form multi-view traffic information perception, so that the perception capability of a key road section is improved, and the technical problems that a front cart can not be accurately perceived in a car scene after shielding, errors in a far-distance scene are large or false targets appear are solved. In a first aspect, an embodiment of the present application provides a traffic perception system based on multi-source information fusion, including a millimeter wave radar and a video camera located in monitoring uprights, each monitoring upright corresponding to two upright perception areas, the traffic perception system including: the radar vehicle detectors are arranged on the road side anti-collision guardrail at intervals, and at least one radar vehicle detector is arranged in each upright rod sensing area and in a handshake area between every two adjacent upright rod sensing areas; The MEC edge computing units are arranged on the monitoring vertical rods one by one, and each MEC edge computing unit is connected with a millimeter wave radar and a video camera corresponding to the monitoring vertical rod, a radar car detector in two vertical rod sensing areas corresponding to the monitoring vertical rod, a radar car detector in a handshake area between the two vertical rod sensing areas corresponding to the monitoring vertical rod, and a radar car detector in a handshake area betwe