Search

CN-122024200-A - Dynamic focus target detection method and system for automatic driving vehicle

CN122024200ACN 122024200 ACN122024200 ACN 122024200ACN-122024200-A

Abstract

The application discloses a method and a system for detecting a dynamic focus target of an automatic driving vehicle, which relate to the technical field of automatic driving, and the method and the system for detecting the dynamic focus target of the automatic driving vehicle are used for carrying out dynamic region priority classification by acquiring vehicle images and speed data, generating an image motion vector field based on motion analysis and combining the vehicle speeds, and outputting dynamic detection frequency planning data, generating a mask of a region to be processed by utilizing a self-adaptive sparse decision module, selectively cutting an image, inputting the mask into a target detection neural network, and finally mapping a coordinate output result, so that intelligent distribution of computing resources is realized, the computing resources can be distributed efficiently, the processing delay is reduced, and the accuracy and the instantaneity of key target detection are improved.

Inventors

  • FANG FUGUI
  • JIN YUANZHI

Assignees

  • 成都锦城学院

Dates

Publication Date
20260512
Application Date
20260410

Claims (10)

  1. 1. A dynamic focus target detection method for an autonomous vehicle, the method comprising: Acquiring a current image frame acquired by an automatic driving vehicle in the driving process and vehicle speed data from a vehicle bus interface; Performing motion analysis based on the current image frame and at least one historical image frame, and outputting an image motion vector field reflecting the motion direction and amplitude of the pixel points; The image motion vector field and the vehicle speed data are used as input to carry out dynamic region priority classification processing and output dynamic detection frequency planning data, wherein the dynamic detection frequency planning data are used for defining a two-dimensional Boolean decision matrix whether different space regions in a current image frame are processed in the current detection or not; The current image frame, the dynamic detection frequency planning data and the historical image frame are taken as inputs and are input to a self-adaptive sparse decision module for processing, and a to-be-processed area mask which is consistent with the size of the current image frame is output; taking the mask of the area to be processed and the current image frame as inputs, performing selective area clipping and splicing processing, and outputting a target candidate area image composed of image contents of the area to be detected marked by the mask; Inputting the target candidate region image into a target detection neural network for processing, and outputting an initial target detection result, wherein the initial target detection result comprises at least one target category and boundary frame coordinates thereof in the target candidate region image; And taking the initial target detection result and the spatial position mapping relation recorded by the to-be-processed area mask as inputs, carrying out coordinate reverse mapping processing, mapping the target boundary frame coordinates to the original coordinate system of the current image frame, and outputting a final target detection result.
  2. 2. The method for detecting a dynamic focus target for an autonomous vehicle according to claim 1, wherein said step of performing a dynamic region prioritization process with said image motion vector field and said vehicle speed data as inputs, and outputting dynamic detection frequency plan data comprises: Dividing the current image frame into a plurality of image subareas, taking the image motion vector field as input, and calculating and outputting a motion change intensity value corresponding to each image subarea; taking the vehicle speed data as input, inquiring a preset vehicle speed-frequency mapping table, and determining and outputting a global detection frequency reference value; Taking the motion change intensity value and the global detection frequency reference value as inputs, performing local modulation processing, and calculating and outputting a corresponding dynamic detection frequency value for each image subarea; And comparing the dynamic detection frequency value of each image subarea with a preset frequency threshold value, and generating and outputting the dynamic detection frequency planning data.
  3. 3. The method for detecting a dynamic focus target for an autonomous vehicle according to claim 2, wherein the step of dividing the current image frame into a plurality of image sub-areas and calculating and outputting a motion variation intensity value corresponding to each of the image sub-areas with the image motion vector field as an input comprises: Extracting all image motion vectors falling within the boundary range of each image subarea; calculating the amplitude of each extracted image motion vector to obtain an amplitude value set of the subarea; carrying out statistical analysis on the amplitude value set to obtain the current frame statistical characteristics of the set, wherein the statistical characteristics comprise at least one of an amplitude average value, an amplitude maximum value and an amplitude standard deviation; reading a historical motion change intensity value corresponding to the image subarea from the continuous multi-frame historical image frames; And carrying out weighted fusion on the current frame statistical characteristics and the historical motion change intensity values, and outputting fusion results to serve as the final motion change intensity values of the image subareas.
  4. 4. The dynamic focus target detection method for an autonomous vehicle according to claim 2, wherein the step of referring to a preset vehicle speed-frequency map, determining and outputting a global detection frequency reference value, taking the vehicle speed data as an input, comprises: judging a vehicle speed section to which the vehicle speed data belong; inquiring a reference frequency value corresponding to the vehicle speed interval from the preset vehicle speed-frequency mapping table, and taking the reference frequency value as the global detection frequency reference value, wherein the preset vehicle speed-frequency mapping table comprises a plurality of discrete vehicle speed intervals and corresponding reference frequency values.
  5. 5. The method for detecting a dynamic focus target for an autonomous vehicle according to claim 2, wherein the step of performing a local modulation process with the motion variation intensity value and the global detection frequency reference value as inputs, and calculating and outputting a corresponding dynamic detection frequency value for each image sub-area comprises: Inquiring a preset mapping relation of motion intensity and modulation coefficient according to the motion change intensity value of each image subarea to obtain and output a corresponding local modulation coefficient; multiplying the global detection frequency reference value by the local modulation coefficient to obtain and output a preliminary frequency value of the image subarea; And applying an upper limit constraint and a lower limit constraint to the preliminary frequency value, and outputting the constrained value as the dynamic detection frequency value of the image subarea, wherein in a preset motion intensity-modulation coefficient mapping relation, a motion change intensity value and a corresponding modulation coefficient are in a direct proportion relation.
  6. 6. The method for detecting a dynamic focus target for an autonomous vehicle according to claim 1, wherein the step of inputting the current image frame, the dynamic detection frequency plan data, and the history image frame as inputs to an adaptive sparse decision module for processing, and outputting a mask of a region to be processed in accordance with a size of the current image frame comprises: Analyzing the dynamic detection frequency planning data, identifying all image subareas which are planned to be detected in the current frame, and marking the corresponding space positions in the mask of the area to be processed as first identification values; Executing space-time content consistency check on image subregions which are planned to be detected in the current frame in the dynamic detection frequency planning data; integrating the first identification value and the identification value updated after verification to form and output the complete area mask to be processed.
  7. 7. The method for dynamic focus target detection for an autonomous vehicle according to claim 6, wherein the step of performing spatiotemporal content consistency check on an image sub-region in said dynamic detection frequency plan data that is planned to be undetected at a current frame comprises: extracting a current image block of the image subarea which does not need to be detected from the current image frame, and extracting a historical image block sequence corresponding to the image subarea from a plurality of frames of historical image frames; Calculating the content difference degree between the current image block and each image block in the historical image block sequence, and aggregating and outputting a comprehensive content difference degree value; If the comprehensive content difference value exceeds a preset consistency threshold, marking the corresponding position update of the image subarea in the to-be-processed area mask as the first identification value; and if the comprehensive content difference value does not exceed the consistency threshold value, marking the corresponding position of the comprehensive content difference value in the mask of the area to be processed as a second identification value.
  8. 8. The method for dynamic focus target detection for an autonomous vehicle according to claim 7, wherein the step of calculating a content difference between the current image block and each image block in the sequence of historical image blocks and aggregating and outputting a composite content difference value comprises: Extracting the characteristics of the current image block and outputting a current characteristic vector; respectively extracting features of each historical image block in the historical image block sequence, and correspondingly outputting a plurality of historical feature vectors; respectively calculating the distance between the current feature vector and each historical feature vector to obtain a plurality of corresponding distance values; and carrying out average calculation on the plurality of distance values, and outputting an average value as the comprehensive content difference degree value.
  9. 9. The dynamic focus target detection method for an autonomous vehicle according to claim 1, characterized in that the method further comprises: Comparing the dynamic detection frequency planning data with the to-be-processed area mask, identifying all image subareas which are planned to be detected in the dynamic detection frequency planning data without detection but marked to be detected in the to-be-processed area mask, and obtaining a difference area set; carrying out space-time distribution statistics on the difference region set, and calculating the aggregation strength of the difference region set in an image frame and the occurrence frequency of the difference region set along with the change of the vehicle speed; Judging whether decision logic for generating the two-dimensional Boolean decision matrix needs to be optimized or not according to the aggregation strength and the occurrence frequency, and generating a corresponding optimization instruction if the decision logic needs to be optimized, and adjusting the decision logic.
  10. 10. A dynamic focus target detection system for an autonomous vehicle, characterized in that the dynamic focus target detection system for an autonomous vehicle comprises a memory, a processor and a dynamic focus target detection program for an autonomous vehicle stored on the memory and executable on the processor, the dynamic focus target detection program for an autonomous vehicle being configured to implement the steps of the dynamic focus target detection method for an autonomous vehicle as claimed in any one of claims 1 to 9.

Description

Dynamic focus target detection method and system for automatic driving vehicle Technical Field The application relates to the technical field of automatic driving, in particular to a method and a system for detecting a dynamic focus target of an automatic driving vehicle. Background In the actual running of an automatic driving system, real-time target detection is used as a core link for guaranteeing driving safety, and the performance of the real-time target detection directly determines the perception capability and response speed of a vehicle to the surrounding environment. The current mainstream technical scheme generally relies on high-performance computing hardware, and performs intensive processing of a full image range on each frame of image continuously captured by a vehicle-mounted camera. However, with the popularity of complex scenes such as urban roads and highways, image data exhibits significant non-uniform characteristics in that a large number of pixels are concentrated in static background areas, such as fixed buildings, road markings, and sky parts, which change weakly or almost constantly in the time dimension, but consume a great deal of computing resources of the processor. At the same time, critical dynamic objects such as pedestrians, non-motor vehicles or vehicles that cut in suddenly, often appear in a sudden manner at specific locations in the field of view, which are short lived and difficult to predict. An inherent disadvantage of the full frame detection mode is that it cannot efficiently identify and prioritize high risk areas, resulting in excessive occupation of computing resources in the information redundancy area, with a significant increase in overall system processing delay. This delay is particularly noticeable under high speed driving conditions, and may cause the system to miss a millisecond critical response window, directly threatening the driving safety. Although there have been studies attempting to introduce motion detection mechanisms or preset fixed-area sampling strategies to reduce computational load, these approaches have significant limitations. The motion detection scheme generally performs region screening only according to local pixel change, fails to perform global coordination by fusing the real-time motion state (such as the running speed) of the vehicle, for example, the detection priority of a front region cannot be automatically improved in a high-speed scene, and the fixed region sampling lacks adaptability to the dynamic change of the scene, and can excessively detect a static region at low speed and easily miss a sudden target of an edge region at high speed. In addition, the contradiction between the detection precision and the resource consumption is difficult to balance in the prior art, and under the condition of complex illumination or shielding, the phenomenon of missed detection or false detection of a key target still can occur. The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present application and is not intended to represent an admission that the foregoing is prior art. Disclosure of Invention The application mainly aims to provide a dynamic focus target detection method and a system for an automatic driving vehicle, which aim to efficiently allocate computing resources, reduce processing delay and improve key target detection accuracy and instantaneity. To achieve the above object, the present application proposes a dynamic focus target detection method for an autonomous vehicle, the method comprising: Acquiring a current image frame acquired by an automatic driving vehicle in the driving process and vehicle speed data from a vehicle bus interface; Performing motion analysis based on the current image frame and at least one historical image frame, and outputting an image motion vector field reflecting the motion direction and amplitude of the pixel points; The image motion vector field and the vehicle speed data are used as input to carry out dynamic region priority classification processing and output dynamic detection frequency planning data, wherein the dynamic detection frequency planning data are used for defining a two-dimensional Boolean decision matrix whether different space regions in a current image frame are processed in the current detection or not; The current image frame, the dynamic detection frequency planning data and the historical image frame are taken as inputs and are input to a self-adaptive sparse decision module for processing, and a to-be-processed area mask which is consistent with the size of the current image frame is output; taking the mask of the area to be processed and the current image frame as inputs, performing selective area clipping and splicing processing, and outputting a target candidate area image composed of image contents of the area to be detected marked by the mask; Inputting the target candidate region image into a target