CN-121980495-A - Intelligent driving multi-sensor fusion data processing system

CN121980495ACN 121980495 ACN121980495 ACN 121980495ACN-121980495-A

Abstract

The invention discloses an intelligent driving multi-sensor fusion data processing system, which relates to the technical field of intelligent driving, solves the problems that the time synchronization of multi-sensor heterogeneous data, the scientific complement of missing data and the dynamic optimization of calculation power distribution cannot be accurately solved, according to the invention, a main body time line is constructed by using the maximum acquisition period as a reference through a standardized period calibration and translation adaptation mechanism of a time characteristic verification end, and the optimal solution of the coincident acquisition points is verified through multiple rounds of translation verification, so that the translation characteristics of the time lines of all sensors are accurately calibrated, and the time asynchronism problem caused by the difference of the acquisition periods of different sensors is effectively eliminated; the linear interpolation processing end is matched to accurately complement the acquired data of the idle period, so that the data integrity and consistency of each sensor in the unified time dimension are ensured, a high-precision data basis is laid for the subsequent fusion processing, and fusion errors caused by time deviation are avoided.

Inventors

HU BIN
ZHAO JUNLAN
ZHANG YUZHEN

Assignees

清华大学苏州汽车研究院（吴江）
清华大学

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (9)

1. An intelligent driving multi-sensor fusion data processing system, comprising: The data characteristic extraction end confirms the acquisition period associated with the corresponding sensor according to the acquisition frequencies associated with different sensors; The time characteristic verification terminal confirms the coincidence progress of a plurality of groups of sensors about the acquired data in a certain period according to a time line front-back translation mode based on different acquisition periods of different sensors, so as to confirm and execute translation characteristics about each sensor; and the linear interpolation processing end confirms the data acquired by each group of sensors, and recognizes the associated data of the corresponding sensor in the acquisition blank period by combining the acquisition time of other sensors, and performs linear interpolation processing.
2. The intelligent driving multi-sensor fusion data processing system according to claim 1, wherein the data feature extraction end confirms the acquisition period in the following specific manner: The acquisition frequencies involved in each set of sensors are labeled P i , where i represents the different sensors; the acquisition period T i associated with the corresponding sensor is confirmed by adopting the T i =1÷P i .
3. The intelligent driving multi-sensor fusion data processing system according to claim 1, wherein the time feature verification terminal confirms the time calibration feature of each sensor in the following specific ways: Generating an acquisition time line related to a corresponding sensor according to the acquisition period related to each group of sensors by taking the current moment as a calibration moment, selecting a maximum value from the acquisition periods related to a plurality of sensors, recording the maximum acquisition period as a standard period, recording the acquisition time line related to the standard period as a standard time line, gradually confirming four groups of acquisition moments from the initial acquisition moment in the standard time line to the back, recording the time lengths related to the five groups of acquisition moments as standard time lengths, and recording the time lines related to the standard time lengths as main body time lines; From the standard time lines related to other sensors, part of time lines with the same standard time length are confirmed from the first initial acquisition time to the rear, and the part of time lines are recorded as verification time lines of the corresponding sensors; Performing front-back translation confirmation on the main body time line and the verification time line, and selecting a standard process from a plurality of front-back translation processes; Confirming the translation characteristic of each group of verification timelines from the standard process, and recalibrating the acquisition state of each group of sensors according to the translation characteristic.
4. An intelligent driving multisensor fusion data processing system of claim 3, wherein the standard process is selected in a manner comprising: Maintaining the time characteristics of the main body time line unchanged, enabling the verification time line to perform time front-back translation confirmation, executing a plurality of groups of translation confirmation processes, marking the coincidence acquisition points associated with each group of translation confirmation processes, wherein the coincidence acquisition points are acquisition points with the same time on different time lines, recording the total number of the coincidence acquisition points marked in the corresponding translation confirmation process as the process characteristics of the corresponding translation confirmation process, selecting the maximum value from the process characteristics associated with different translation confirmation processes, and recording the translation confirmation process associated with the maximum value as the standard process.
5. The intelligent driving multi-sensor fusion data processing system according to claim 3, wherein the specific way for the linear interpolation processing end to confirm the associated data of different sensors in the acquisition idle period is as follows: confirming the data acquired by each group of sensors, identifying the acquisition time, confirming whether the acquired data exist in other sensors at the acquisition time, if so, not needing any processing, and if not, carrying out linear interpolation processing on the appointed sensor; Marking the acquisition time as C 0 , marking the acquisition time when the acquisition data exists before the designated sensor is C 0 as C 1 , marking the acquisition time when the acquisition data exists after the designated sensor is C 0 as C 2 , marking the acquisition data associated with C1 as SJ 1 , and marking the acquisition data associated with C2 as SJ 2 ; The method adopts the following steps: And confirming the associated data SJ 0 corresponding to the acquisition time C 0 , and performing linear interpolation processing based on the confirmed associated data SJ 0 , so that the follow-up data synchronization confirmation is facilitated.
6. The intelligent driving multi-sensor fusion data processing system of claim 1, further comprising: the feature fusion processing end confirms fusion features related to each group of acquisition data from historical data aiming at a plurality of groups of acquisition data related to the same acquisition time, confirms fusion weights related to different acquisition data based on the fusion features, and transmits the confirmed fusion weights to the dynamic weight distribution end.
7. The intelligent driving multi-sensor fusion data processing system according to claim 6, wherein the feature fusion processing end performs a confirmation process for different collected data fusion features, including: the data types of a plurality of groups of acquired data at the same acquisition time are marked as undetermined types, processing data related to the undetermined types are identified from historical data, fusion data of different batches are extracted from the processing data, execution computing power and generated fusion average speed are confirmed from the fusion data, a processing mode of fusion average speed/execution computing power = fusion characteristic is adopted, the fusion characteristic of the fusion data is confirmed, the maximum value is extracted from fusion characteristics related to different fusion data, and the execution computing power related to the maximum value is marked as the fusion characteristic related to the undetermined types.
8. The intelligent driving multi-sensor fusion data processing system according to claim 7, wherein the characteristic fusion processing end confirms fusion weights of different collected data in the following specific ways: According to different fusion characteristics associated with different undetermined types, carrying out ratio processing on a plurality of groups of fusion characteristics, confirming a plurality of groups of characteristic ratio columns associated with different undetermined types, identifying inherent calculation forces set in a current intelligent driving system, uniformly distributing the inherent calculation forces according to a ratio arrangement mode of the characteristic ratio columns, confirming distribution calculation forces associated with each undetermined type, directly extracting the associated calculation forces with the values consistent with the fusion characteristics from the inherent calculation forces as the distribution calculation forces associated with the corresponding undetermined types if the distribution calculation forces associated with each undetermined type exceed the fusion characteristics, and keeping the confirmed distribution calculation forces unchanged if the distribution calculation forces associated with each undetermined type do not exceed the fusion characteristics; And recording the confirmed distribution calculation force of each pending type as the fusion weight of the corresponding acquired data, and transmitting the confirmed fusion weight into the dynamic weight distribution terminal.
9. The intelligent driving multi-sensor fusion data processing system according to claim 6, wherein the dynamic weight distribution terminal distributes inherent calculation forces existing in the intelligent driving system according to different fusion weights associated with different acquired data to complete the unified fusion processing process of the different acquired data.

Description

Intelligent driving multi-sensor fusion data processing system Technical Field The invention relates to the technical field of intelligent driving, in particular to an intelligent driving multi-sensor fusion data processing system. Background The intelligent driving technology has the core that accurate cognition on complex traffic environments is realized through multi-sensor cooperative sensing, and the characteristics of heterogeneous acquisition frequency, data format and time reference commonly exist due to principle differences of sensors such as a laser radar, a camera, a millimeter wave radar and the like, so that fundamental challenges are brought to multi-source data fusion. Currently, most intelligent driving multi-sensor data processing schemes exist with significantly short plates: firstly, a time synchronization mechanism is crude, a crystal oscillator clock of a sensor or a simple time stamp of the sensor is aligned, the difference of acquisition periods of different sensors is not fully considered, data time asynchronization is easily caused, and fusion errors are introduced; secondly, aiming at the problem of data missing in the sensor acquisition idle period, a precise complement strategy is lacking, data fracture or interpolation distortion often occurs, and the continuity and the integrity of fusion data are affected; Thirdly, the calculation force distribution is lack of scientificity, the calculation force is distributed by adopting fixed weight or experience value, the actual processing requirement of the data type is not combined with the dynamic adaptation, the redundancy of the calculation force of partial data processing exists, the insufficient key data fusion rate can be caused, and the severe requirement of intelligent driving on real-time performance is difficult to meet; in addition, the core processing logic (such as time calibration and weight confirmation) of the existing scheme lacks standardized and floor-capable execution flow, the flexibility of adapting to different sensors is poor, a large amount of manual debugging is needed during deployment, and the application cost and the floor difficulty are increased. Therefore, a set of processing system capable of accurately solving the time synchronization of heterogeneous data of multiple sensors, scientifically supplementing missing data and dynamically optimizing calculation power distribution is developed, and the processing system becomes a key requirement for improving the perception precision and instantaneity of an intelligent driving environment and promoting the large-scale application of an intelligent driving technology. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an intelligent driving multi-sensor fusion data processing system, which solves the problems that the time synchronization, the scientific data complement and the dynamic optimization of calculation force distribution of multi-sensor heterogeneous data cannot be accurately solved. In order to achieve the purpose, the intelligent driving multi-sensor fusion data processing system is realized by the following technical scheme that the intelligent driving multi-sensor fusion data processing system comprises: The data characteristic extraction end confirms the acquisition period associated with the corresponding sensor according to the acquisition frequencies associated with different sensors, and the specific mode is as follows: The acquisition frequencies involved in each set of sensors are labeled P i, where i represents the different sensors; Confirming the acquisition period T i associated with the corresponding sensor by adopting T i=1÷Pi; The time characteristic verification terminal confirms the coincidence progress of a plurality of groups of sensors about the acquired data in a certain period according to a time line front-back translation mode based on different acquisition periods of different sensors, so as to confirm and execute translation characteristics about each sensor, and the specific mode is as follows: Generating an acquisition time line related to a corresponding sensor according to the acquisition period related to each group of sensors by taking the current moment as a calibration moment, selecting a maximum value from the acquisition periods related to a plurality of sensors, recording the maximum acquisition period as a standard period, recording the acquisition time line related to the standard period as a standard time line, gradually confirming four groups of acquisition moments from the initial acquisition moment in the standard time line to the back, recording the time lengths related to the five groups of acquisition moments as standard time lengths, and recording the time lines related to the standard time lengths as main body time lines; From the standard time lines related to other sensors, part of time lines with the same standard time length are confirmed from the first initi