CN-121977529-A - Map construction method, device, equipment and storage medium

CN121977529ACN 121977529 ACN121977529 ACN 121977529ACN-121977529-A

Abstract

The application provides a map construction method, a map construction device, map construction equipment and a storage medium. The method comprises the steps of determining a positioning state of a current frame based on current frame data input by a sensor and a map set containing a plurality of sub-maps, managing the map set based on the positioning state, including updating or determining a current active map, executing local optimization on the current active map based on the managed map set, fusing the optimized current active map with other sub-maps in the map set, and executing global optimization on the fused result to generate a global consistent target map. By adopting the method, the accuracy, the continuity and the global consistency of the map construction are obviously enhanced.

Inventors

CHENG XIANGQIAN
Cao Rongchuan
Wang Dainan
ZHANG SONGLIN

Assignees

中国第一汽车股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260202

Claims (10)

1. A method of map construction, the method comprising: Determining a positioning state of the current frame based on the current frame data input by the sensor and a map set containing a plurality of sub-maps; based on the positioning state, managing the atlas, including updating or determining a current active map; Performing local optimization on the current active map based on the managed map set; and fusing the optimized current active map with other sub-maps in the map set, and executing global optimization on the fused result to generate a globally consistent target map.
2. The method of claim 1, wherein the sensor-entered current frame data is obtained based on a sensor state for which initialization has been completed, wherein the initialization includes a vision-inertial initialization process comprising: based on the pure visual information of the continuous image frames, performing visual motion restoration to obtain a visual motion trail for subsequent optimization; Determining initial values of scale factors, gravity directions, sensor bias and speed through pure inertia optimization based on the vision motion track and pre-integral data of an inertia measurement unit; And determining final initialization parameters of the system through joint optimization of visual and inertial data based on the initial values, wherein the initialization parameters are used for configuring the sensor states to provide physical dimensions and spatial references for determining the positioning states based on the sensor inputs in claim 1.
3. The method of claim 1, wherein determining the positioning state of the current frame based on the current frame data entered by the sensor and a map set comprising a plurality of sub-maps comprises: performing feature matching on the current frame and the candidate sub-map in the map set to obtain a matching result; calculating the pose of the current frame relative to the candidate sub-map through motion estimation based on the matching result; And judging whether positioning is successful or not based on the matching result and the re-projection error corresponding to the calculated pose, and if so, determining the candidate sub-map as the current active map.
4. The method of claim 1, wherein the managing the atlas based on the localization status includes updating or determining a current active map, comprising: When the positioning state is successful, adding key frames and map points generated based on the current frame to the current active map; and when the positioning state is failure, creating a new sub-map in the map set, and determining the new sub-map as the current active map.
5. The method of claim 1, wherein the performing local optimization on the current active map based on the managed atlas comprises: Constructing a local optimization problem in the current active map based on the observation relation between the key frames added to the map and the map points; And executing a local beam method adjustment to solve the local optimization problem so as to realize the local optimization.
6. The method of claim 1, wherein fusing the optimized current active map with other sub-maps in the map set and performing global optimization on the fused result comprises: Determining association relation and relative transformation between the optimized current active map and other sub-maps in the map set through scene recognition; Based on the association relation and the relative transformation, aligning and fusing the optimized current active map with local areas of other sub-maps; And executing global beam method adjustment or pose diagram optimization on the aligned and fused map to realize the global optimization.
7. The method according to claim 1 or 6, wherein the fusing the optimized current active map with other sub-maps in the map set comprises: Determining a fusion window, wherein the fusion window comprises a local area from the optimized current active map and another sub-map to be fused in the map set; And executing welding beam method adjustment in the fusion window so as to simultaneously optimize the positions of the key frame poses and map points from the two sub-maps and realize the fusion.
8. A map construction apparatus, characterized in that the apparatus comprises: The positioning state determining module is used for determining the positioning state of the current frame based on the current frame data input by the sensor and a map set containing a plurality of sub-maps; The map set updating module is used for managing the map set based on the positioning state and comprises the steps of updating or determining a current active map; The local optimization module is used for executing local optimization on the current active map based on the managed map set; and the global optimization module is used for fusing the optimized current active map with other sub-maps in the map set, and executing global optimization on the fused result to generate a globally consistent target map.
9. A computer device comprising a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory in communication via the bus when the computer device is in operation, the machine-readable instructions when executed by the processor performing the steps of the mapping method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, performs the steps of the map construction method according to any of claims 1 to 7.

Description

Map construction method, device, equipment and storage medium Technical Field The present application relates to the field of computer technologies, and in particular, to a map construction method, apparatus, device, and storage medium. Background The visual synchronous positioning and map construction (SLAM) technology is a core support technology in the fields of robot navigation, automatic driving, augmented Reality (AR) and the like, and the core target is environment data acquired through a sensor under the condition of no priori environment information, and the pose of the equipment is calculated in real time and a global consistent environment map is constructed. Along with the expansion of application scenes, the map construction method has the advantages that the map construction method has higher requirements on the map construction precision, robustness and map multiplexing capability of an SLAM system from indoor short-distance operation to outdoor large-range movement and from single task session to multi-period cross-scene operation, and the integrity and consistency of map construction have become key factors restricting the technology to fall to the ground. In the prior art, the mainstream map construction scheme mostly adopts a single map architecture based on key frame and beam method adjustment (BA) optimization, typically representing an ORB-SLAM2 system. The system extracts image features and matches key frames through tracking threads, the local mapping threads optimize the pose of the key frames and map point coordinates, the closed loop detection threads modify accumulated drift through feature matching and pose optimization, and finally, a single continuous environment map is constructed, so that a good application effect is achieved in scenes with small and medium ranges and short time sequences. However, the prior art has the obvious defect that the adopted single map architecture lacks an effective management and fusion mechanism for the multi-session map. When the system loses tracking due to shielding, rapid movement or illumination mutation in the tracking process, a brand new map is required to be reinitialized and constructed, the previously constructed historical map data cannot be reused, even if the system enters the same area again later, the new map and the historical map are difficult to realize accurate alignment and fusion, the maps constructed in different sessions are independent and cannot be unified, data redundancy is caused, accumulated drift is generated due to repeated map construction, and global consistency and application reliability of map construction in a large-scale and multi-period scene are seriously influenced. Disclosure of Invention In view of the above, the present application aims to provide a map construction method, apparatus, device and storage medium, which can significantly enhance the accuracy, continuity and global consistency of map construction. In a first aspect, an embodiment of the present application provides a map construction method, where the method includes: Determining a positioning state of the current frame based on the current frame data input by the sensor and a map set containing a plurality of sub-maps; based on the positioning state, managing the atlas, including updating or determining a current active map; Performing local optimization on the current active map based on the managed map set; and fusing the optimized current active map with other sub-maps in the map set, and executing global optimization on the fused result to generate a globally consistent target map. Optionally, the current frame data of the sensor input is obtained based on the sensor state that has been initialized, the initializing comprising a vision-inertia initialization process comprising: based on the pure visual information of the continuous image frames, performing visual motion restoration to obtain a visual motion trail for subsequent optimization; Determining initial values of scale factors, gravity directions, sensor bias and speed through pure inertia optimization based on the vision motion track and pre-integral data of an inertia measurement unit; And determining final initialization parameters of the system through joint optimization of visual and inertial data based on the initial values, wherein the initialization parameters are used for configuring the sensor states to provide physical dimensions and spatial references for determining the positioning states based on the sensor inputs in claim 1. Optionally, the determining the positioning state of the current frame based on the current frame data input by the sensor and a map set including a plurality of sub-maps includes: performing feature matching on the current frame and the candidate sub-map in the map set to obtain a matching result; calculating the pose of the current frame relative to the candidate sub-map through motion estimation based on the matching result; And judging