WO2022021872A1 - 目标检测方法及装置、电子设备和存储介质 - Google Patents

目标检测方法及装置、电子设备和存储介质 Download PDF

Info

Publication number
WO2022021872A1
WO2022021872A1 PCT/CN2021/078481 CN2021078481W WO2022021872A1 WO 2022021872 A1 WO2022021872 A1 WO 2022021872A1 CN 2021078481 W CN2021078481 W CN 2021078481W WO 2022021872 A1 WO2022021872 A1 WO 2022021872A1
Authority
WO
WIPO (PCT)
Prior art keywords
frame
point cloud
target
cloud data
detection
Prior art date
Application number
PCT/CN2021/078481
Other languages
English (en)
French (fr)
Inventor
鲍虎军
周晓巍
孙佳明
谢一鸣
张思宇
Original Assignee
浙江商汤科技开发有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 浙江商汤科技开发有限公司 filed Critical 浙江商汤科技开发有限公司
Priority to KR1020227003199A priority Critical patent/KR20220027202A/ko
Priority to JP2022505272A priority patent/JP2022546201A/ja
Publication of WO2022021872A1 publication Critical patent/WO2022021872A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/64—Three-dimensional objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00—Indexing scheme for image generation or computer graphics
    • G06T2210/12—Bounding box
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Definitions

  • the present disclosure relates to the field of computer technology, and in particular, to a target detection method and device, an electronic device, and a storage medium.
  • Object detection is a very important task in computer vision, which can estimate the pose, scale and other information of objects (such as people or objects) within the field of view through the input data of sensors.
  • the target detection method usually processes the input of each frame separately, resulting in poor detection accuracy.
  • the present disclosure proposes a technical solution for target detection.
  • a target detection method comprising: performing target detection on point cloud data of frame t of a target scene, and determining a first candidate frame of the target in the point cloud data of frame t, where t is an integer greater than 1; according to the point cloud data of the t frame, the first candidate frame and the prediction candidate frame for the target in the point cloud data of the t frame, determine the first frame of the point cloud data of the t frame.
  • the detection result, the first detection result includes the first detection frame of the target in the t-th frame of point cloud data, wherein the prediction candidate frame is based on the t-1 frame point before the t-th frame of point cloud data
  • the detection results of cloud data are predicted. In this way, the detection frame is predicted by using multi-frame point cloud data, which can improve the accuracy of target detection.
  • performing target detection on the point cloud data of the t-th frame of the target scene and determining the first candidate frame of the target in the point cloud data of the t-th frame includes: according to the t-th frame The predicted probability map of the target in the point cloud data, the point cloud data of the t-th frame is divided into a first area with a target, a second area without a target, and a third area where it is not determined whether there is a target; Target detection is performed on an area and the third area, and a first candidate frame of the target in the point cloud data of the t-th frame is determined.
  • the amount of point cloud data processed for object detection can be reduced, and the detection speed can be improved.
  • the method further includes: acquiring a second detection result of the point cloud data of the t-1 frame before the point cloud data of the t frame; according to the point cloud data of the t frame The second detection result of the point cloud data of the previous t-1 frame is corrected to the first detection result of the point cloud data of the t-th frame to determine the second detection result of the point-cloud data of the t-th frame. If, by further correcting the first detection result, the accuracy of the first detection result can be further improved.
  • the method further includes: according to the second detection result of the point cloud data of the t-1 frame before the point cloud data of the t-th frame, for the point cloud data of the t-th frame The motion state of the target is predicted, and the prediction candidate frame of the target in the point cloud data of the t-th frame is determined. In this way, the prediction candidate frame of the target in the point cloud data of the t-th frame is predicted based on the previous multi-frame point cloud data, which can improve the prediction accuracy.
  • the method further includes: according to the prediction candidate frame of the target in the point cloud data of the t-th frame, and the point cloud data of the t-1-th frame, for the point cloud of the t-1-th frame
  • the predicted probability map of the target in the cloud data is updated, and the predicted probability map of the target in the point cloud data of the t-th frame is determined.
  • the probability of the occurrence of the target can be further predicted, so as to improve the accuracy of the finally obtained prediction probability map.
  • the performing target detection on the first area and the third area to determine the first candidate frame of the target in the point cloud data of the t-th frame includes: detecting the target in the t-th frame of point cloud data Perform feature extraction on the point cloud data of an area and the third area to obtain the first point cloud feature; perform target detection on the first point cloud feature to determine the second candidate of the target in the point cloud data of the t-th frame frame; according to the confidence of each second candidate frame, determine a preset number of first candidate frames from the second candidate frame.
  • the sampled point cloud data is input into the feature extraction network for processing to obtain the first point cloud feature, which can further improve the detection speed.
  • the t-th frame is determined according to the t-th frame of point cloud data, the first candidate frame, and a prediction candidate frame for the target in the t-th frame of point cloud data
  • the first detection result of the point cloud data includes: respectively extending the prediction candidate frames of each target in the point cloud data of the t-th frame to determine the third candidate frame of each target; comparing the third candidate frame and the The first candidate frames are respectively matched to determine the target corresponding to each first candidate frame; according to the first candidate frame and the first area point cloud data corresponding to the area where the first candidate frame is located, and the third The candidate frame and the second area point cloud data corresponding to the area where the third candidate frame is located, perform candidate frame fusion on each target in the t-th frame of point cloud data to obtain the t-th frame of point cloud data.
  • the first detection frame of each target In this way, the probability of matching with the first candidate frame can be increased, so as to improve the stability of the detection result.
  • the matching of the third candidate frame and the first candidate frame respectively, and determining the target corresponding to each first candidate frame includes: respectively determining that each third candidate frame and the The intersection ratio between each first candidate frame; the third candidate frame whose intersection ratio with the first candidate frame is greater than or equal to the intersection ratio threshold is determined as the third candidate frame that matches the first candidate frame; The target corresponding to the third candidate frame matching the first candidate frame is determined as the target corresponding to the first candidate frame. In this way, by determining a candidate frame with a large intersection ratio as a matching candidate frame, the accuracy of the prediction candidate frame can be improved.
  • each second detection result includes a second detection frame of the target, and according to the second detection result of the t-1 frame of point cloud data before the t-th frame of point cloud data, Correcting the first detection result of the point cloud data of the t-th frame, and determining the second detection result of the point-cloud data of the t-th frame, includes: determining the detection frame set of the first target, and the first target is the Any target in the t-th frame of point cloud data, the detection frame set of the first target includes the second detection frame of the first target in the second detection result of the t-1 frame of point cloud data, and all The first detection frame of the first target in the first detection result of the point cloud data of the t-th frame; for any detection frame in the detection frame set of the first target, compare the detection frame set with the detection frame The detection frame whose error between the frames is less than or equal to the error threshold is determined as the inner point frame of the detection frame; the third detection frame with the largest number of inner point frames is determined from the detection frame
  • the method further includes: according to the second detection result of the point cloud data of the t-1 frame and the second detection result of the point cloud data of the t frame, for the t+th frame
  • the motion state of the target in the 1 frame of point cloud data is predicted, and the prediction candidate frame of the target in the t+1 th frame of point cloud data is determined.
  • the motion state of the target in the point cloud data of the next frame can be predicted, and then the detection of the target in the point cloud data of the next frame can be realized.
  • the method further includes: according to the prediction candidate frame of the target in the t+1 th frame of point cloud data and the t th frame of point cloud data, performing an analysis on the t th frame of point cloud data
  • the predicted probability map of the target in the t+1 frame is updated to determine the predicted probability map of the target in the t+1th frame of point cloud data.
  • performing target detection on the point cloud data of the t-th frame of the target scene, and determining the first candidate frame of the target in the point cloud data of the t-th frame includes: Perform feature extraction on the point cloud data to obtain a second point cloud feature; perform target detection on the second point cloud feature to determine the fourth candidate frame of the target in the t-th frame of point cloud data; Confidence, determining a preset number of first candidate frames from the fourth candidate frame. In this way, target detection can be performed on the point cloud data of the t frame without performing region division on the point cloud data of the t frame.
  • the first detection result further includes the category of the target in the point cloud data of the t-th frame, and the target type according to the point cloud data of the t-th frame, the first candidate frame and the target
  • the prediction candidate frame of the target in the point cloud data of the t-th frame, and determining the first detection result of the point cloud data of the t-th frame including: according to the third area point corresponding to the area where the first detection frame of the second target is located Cloud data, classify the second target, determine the category of the second target, and the second target is any target in the point cloud data of the t-th frame.
  • the target scene includes an indoor scene
  • the target in the t-th frame of point cloud data includes an object
  • the first detection frame of the target in the t-th frame of point cloud data includes a three-dimensional area frame .
  • a target detection device comprising:
  • a first detection module configured to perform target detection on the point cloud data of the t-th frame of the target scene, and determine the first candidate frame of the target in the point cloud data of the t-th frame, where t is an integer greater than 1;
  • the second detection module is configured to determine the t-th frame of point cloud data according to the t-th frame of point cloud data, the first candidate frame and the predicted candidate frame for the target in the t-th frame of point cloud data.
  • the first detection result, the first detection result includes the first detection frame of the target in the point cloud data of the t-th frame,
  • the prediction candidate frame is predicted according to the detection result of the point cloud data of the t-1 frame before the point cloud data of the t frame.
  • the first detection module includes: a region division sub-module, configured to divide the point cloud data of the t-th frame according to the predicted probability map of the target in the point-cloud data of the t-th frame is a first area with a target, a second area without a target, and a third area where it is not determined whether there is a target; a first detection submodule is configured to perform target detection on the first area and the third area, Determine the first candidate frame of the target in the point cloud data of the t-th frame.
  • the apparatus further includes: a correction module configured to acquire the second detection result of the point cloud data of the t-1 frame before the point cloud data of the t-th frame; The second detection result of the t-1 frame of point cloud data before the t-th frame of point cloud data, the first detection result of the t-th frame of point cloud data is corrected, and the second detection result of the t-th frame of point cloud data is determined. Test results.
  • the apparatus further includes: a first motion prediction module configured to, according to the second detection result of the point cloud data of the t-1 frame before the point cloud data of the t frame, perform a The motion state of the target in the point cloud data of the t-th frame is predicted, and the prediction candidate frame of the target in the point cloud data of the t-th frame is determined.
  • a first motion prediction module configured to, according to the second detection result of the point cloud data of the t-1 frame before the point cloud data of the t frame, perform a The motion state of the target in the point cloud data of the t-th frame is predicted, and the prediction candidate frame of the target in the point cloud data of the t-th frame is determined.
  • the apparatus further includes: a first probability map update module, configured to predict candidate frames of the target in the t-th frame of point cloud data, and the t-1-th frame of point cloud data, The predicted probability map of the target in the t-1th frame of point cloud data is updated to determine the predicted probability map of the target in the t-th frame of point cloud data.
  • a first probability map update module configured to predict candidate frames of the target in the t-th frame of point cloud data, and the t-1-th frame of point cloud data, The predicted probability map of the target in the t-1th frame of point cloud data is updated to determine the predicted probability map of the target in the t-th frame of point cloud data.
  • the first detection sub-module is configured to: perform feature extraction on the point cloud data of the first area and the third area to obtain a first point cloud feature;
  • the first point cloud feature is used for target detection, and the second candidate frame of the target in the point cloud data of the t-th frame is determined; according to the confidence of each second candidate frame, a preset number of The first candidate box.
  • the second detection module includes: a candidate frame expansion sub-module, configured to expand the prediction candidate frames of each target in the point cloud data of the t-th frame respectively, and determine the th Three candidate frames; a candidate frame matching submodule, configured to match the third candidate frame and the first candidate frame respectively, to determine the target corresponding to each first candidate frame; the candidate frame fusion submodule, configured to The first candidate frame and the first region point cloud data corresponding to the region where the first candidate box is located, and the third candidate box and the second region point cloud data corresponding to the region where the third candidate box is located , performing candidate frame fusion on each target in the t-th frame of point cloud data, to obtain a first detection frame of each target in the t-th frame of point cloud data.
  • the candidate frame matching sub-module is configured to: respectively determine the intersection ratio between each third candidate frame and each first candidate frame; compare the intersection ratio with the first candidate frame The third candidate frame that is greater than or equal to the intersection ratio threshold is determined as the third candidate frame that matches the first candidate frame; the target corresponding to the third candidate frame that matches the first candidate frame is determined to match the The target corresponding to the first candidate frame.
  • each second detection result includes a second detection frame of the target
  • the correction module includes: a set determination submodule configured to determine a detection frame set of a first target, the first target is any target in the t-th frame of point cloud data, and the detection frame set of the first target includes the second detection frame of the first target in the second detection result of the t-1 frame of point cloud data, and the first detection frame of the first target in the first detection result of the point cloud data of the t-th frame;
  • the inner point frame determination submodule is configured to be any detection frame in the detection frame set for the first target, A detection frame whose error between the detection frame set and the detection frame is less than or equal to an error threshold is determined as the inner point frame of the detection frame;
  • the detection frame selection sub-module is configured to select the detection frame from the first target Determine the third detection frame with the largest number of in-point frames in the detection frame set of the The second detection frame of the first target in the point cloud data of the t-th frame.
  • the apparatus further includes: a second motion prediction module, configured to be based on the second detection result of the point cloud data of the t-1 frame, and the second detection result of the point cloud data of the t frame The second detection result is to predict the motion state of the target in the t+1th frame of point cloud data, and determine the prediction candidate frame of the target in the t+1th frame of point cloud data.
  • a second motion prediction module configured to be based on the second detection result of the point cloud data of the t-1 frame, and the second detection result of the point cloud data of the t frame The second detection result is to predict the motion state of the target in the t+1th frame of point cloud data, and determine the prediction candidate frame of the target in the t+1th frame of point cloud data.
  • the apparatus further includes: a second probability map update module, configured to predict candidate frames of the target in the t+1 th frame of point cloud data, and the t th frame of point cloud data, The predicted probability map of the target in the t-th frame of point cloud data is updated, and the predicted probability map of the target in the t+1-th frame of point cloud data is determined.
  • a second probability map update module configured to predict candidate frames of the target in the t+1 th frame of point cloud data, and the t th frame of point cloud data, The predicted probability map of the target in the t-th frame of point cloud data is updated, and the predicted probability map of the target in the t+1-th frame of point cloud data is determined.
  • the first detection module includes: a feature extraction sub-module configured to perform feature extraction on the point cloud data of the t-th frame to obtain second point cloud features; the second detection sub-module, is configured to perform target detection on the second point cloud feature, and determine the fourth candidate frame of the target in the point cloud data of the t-th frame; the selection sub-module is configured to, according to the confidence of each fourth candidate frame, select the target frame from the The fourth candidate frame determines a preset number of first candidate frames.
  • the first detection result further includes the category of the target in the point cloud data of the t-th frame
  • the second detection module includes: a classification sub-module configured to The third area point cloud data corresponding to the area where the first detection frame is located, classify the second target, determine the category of the second target, and the second target is any point cloud data in the t-th frame. a goal.
  • the target scene includes an indoor scene
  • the target in the t-th frame of point cloud data includes an object
  • the first detection frame of the target in the t-th frame of point cloud data includes a three-dimensional area frame .
  • an electronic device comprising: a processor; a memory configured to store instructions executable by the processor; wherein the processor is configured to invoke the instructions stored in the memory to execute the above method.
  • a computer-readable storage medium having computer program instructions stored thereon, the computer program instructions implementing the above method when executed by a processor.
  • the first candidate frame of the target in the point cloud data of the t-th frame can be detected; the predicted candidate frame obtained by the historical detection results is used to modify the first candidate frame, and the first candidate frame of the point cloud data of the t-th frame is obtained. detection results, thereby improving the accuracy of target detection.
  • FIG. 1A shows a flowchart of a target detection method according to an embodiment of the present disclosure.
  • FIG. 1B shows a schematic diagram of a network architecture of a target detection method according to an embodiment of the present disclosure
  • FIG. 2 shows a schematic diagram of a processing procedure of a target detection method according to an embodiment of the present disclosure.
  • Figure 3a shows a schematic diagram of an image of a target scene.
  • Figure 3b shows a schematic diagram of the detection result of the target.
  • FIG. 4 shows a block diagram of a target detection apparatus according to an embodiment of the present disclosure.
  • FIG. 5 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
  • FIG. 6 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
  • FIG. 1A shows a flowchart of a target detection method according to an embodiment of the present disclosure. As shown in FIG. 1A , the target detection method includes:
  • step S11 target detection is performed on the t-th frame of point cloud data of the target scene, and the first candidate frame of the target in the t-th frame of point cloud data is determined;
  • t is an integer greater than 1.
  • step S12 according to the point cloud data of the t frame, the first candidate frame and the prediction candidate frame for the target in the point cloud data of the t frame, determine the first frame of the point cloud data of the t frame.
  • a detection result the first detection result includes the first detection frame of the target in the point cloud data of the t-th frame;
  • the prediction candidate frame is predicted according to the detection result of the point cloud data of the t-1 frame before the point cloud data of the t frame.
  • the target detection method may be performed by an electronic device such as a terminal device or a server
  • the terminal device may be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless Telephone, personal digital assistant (Personal Digital Assistant, PDA), handheld device, computing device, vehicle-mounted device, wearable device, etc.
  • the method can be implemented by the processor calling the computer-readable instructions stored in the memory.
  • the method may be performed by a server.
  • the target scene may include indoor scenes such as shopping malls, hospitals, and exhibition halls, and may also include outdoor scenes such as transportation hubs and city streets.
  • the target scene may include various categories of targets, such as objects, signs, buildings, pedestrians, vehicles, etc. The present disclosure does not limit the types of target scenarios and categories of targets.
  • the sensing data of the target scene may be collected by the sensing device, so as to analyze the target in the sensing data.
  • the sensing device may include, for example, a lidar, a Red Green Blue-Depth (RGB-D) acquisition device, etc.
  • the collected sensing data may include point cloud data, RGB -D image data etc.
  • the present disclosure does not limit the types of sensing devices and the specific types of collected sensing data.
  • multiple frames of sensing data of the target scene may be continuously collected, and target detection is performed on each frame of sensing data in sequence through an electronic device.
  • the sensing data is point cloud data, it can be processed directly; if the sensing data is RGB-D image data, the RGB-D image data can be back-projected and converted to obtain point cloud data for processing.
  • target detection may be performed directly on the first frame of point cloud data in step S11 to obtain the first frame of the target in the first frame of point cloud data. candidate frame; and directly fuse the first candidate frame in step S12 to obtain the first detection frame of the target of the first frame of point cloud data.
  • target detection may be performed on the t-th frame of point cloud data in step S11 to determine the t-th frame point.
  • the information of the first candidate frame may include information such as three-dimensional coordinates (x 0 , y 0 , z 0 ), length, width, height, and rotation angle of the center point of the first candidate frame.
  • the process of target detection can be implemented through a pre-trained target detection network
  • the target detection network may include, for example, a convolutional neural network (Convolutional Neural Networks, CNN) and a region generation network (Region Proposal Network, RPN), the present disclosure does not limit the specific network structure of the target detection network.
  • CNN convolutional Neural Networks
  • RPN Region Proposal Network
  • the detection result of the point cloud data of the t-1th frame before the t-th frame of point cloud data can be detected according to the As a result, the positions of the detected targets in the point cloud data of the previous t-1 frames are predicted in the point cloud data of the t frame, and the predicted candidate frames of these targets in the point cloud data of the t frame are obtained.
  • the target corresponding to each first candidate frame may be determined. For example, according to the intersection ratio of each first candidate frame and each prediction candidate frame, the first candidate frame and the prediction candidate frame are matched; for the first candidate frame in which there is a matching prediction candidate frame, the matching prediction candidate frame The corresponding target is determined as the target corresponding to the first candidate frame; for the first candidate frame for which there is no matching prediction candidate frame, it is determined that the first candidate frame corresponds to a new target.
  • the regional point cloud corresponding to the predicted candidate frame and the predicted candidate frame of the target can be obtained. data, and perform candidate frame fusion processing to determine the actual detection frame of the target (which may be referred to as the first detection frame).
  • the candidate frame fusion can be achieved through a pre-trained fusion network, that is, the first candidate frame of the target and the regional point cloud data corresponding to the first candidate frame, and the predicted candidate frame and prediction of the target.
  • the regional point cloud data corresponding to the candidate frame is input into the fusion network for processing, and the first detection frame of the target is output.
  • the fusion network may include, for example, a regional convolutional neural network (Region Convolutional Neural Networks, RCNN), and the present disclosure does not limit the specific network structure of the fusion network.
  • the first detection result of the point cloud data of the t-th frame can be obtained, and the first detection result includes the point cloud of the t-th frame The first detection frame of each target in the data.
  • the first candidate frame of the target in the point cloud data of the t-th frame can be detected; the predicted candidate frame obtained by the historical detection results is used to modify the first candidate frame, and the first candidate frame of the point cloud data of the t-th frame is obtained. detection results, thereby improving the accuracy of target detection.
  • step S11 may include:
  • the point cloud data of the t-th frame is divided into a first area with a target, a second area without a target, and a third area where it is not determined whether there is a target area;
  • the prediction candidate frame for the target in the point cloud data of the t-th frame can be predicted according to the detection result of the point cloud data of the previous t-1 frame.
  • the prediction candidate frame the probability of the target appearing at each position of the point cloud data of the t-th frame can be predicted, and the predicted probability map of the target in the point cloud data of the t-th frame is obtained.
  • a first probability threshold and a second probability threshold may be preset, and the second probability threshold is smaller than the first probability threshold.
  • the probability of the target appearing at the position is greater than the first probability threshold, it can be considered that there is a target at the position; if the probability of the target appearing at the position is less than the second probability threshold, it can be considered that the position does not exist. target; if the probability of the target appearing at the position is between the first probability threshold and the second probability threshold, it is not certain whether there is a target at this position, such as a position that has not been detected, or a position that has been detected but still not determined whether there is a target .
  • the present disclosure does not limit the specific values of the first probability threshold and the second probability threshold.
  • the point cloud data of the t frame can be divided into the first area where the target exists based on the first probability threshold and the second probability threshold , a second area where there is no target, and a third area where it is not determined whether there is a target.
  • target detection may not be performed on the point cloud data of the second area. That is, target detection is performed on the first area and the third area of the t-th frame of point cloud data, and the first candidate frame of the target in the t-th frame of point cloud data is determined.
  • the data volume of the point cloud data processed by the target detection can be reduced, and the detection speed can be improved.
  • the step of performing target detection on the first area and the third area of the point cloud data of the t-th frame, and determining the first candidate frame of the target in the point-cloud data of the t-th frame may include:
  • a preset number of first candidate frames are determined from the second candidate frames.
  • the point cloud data of the first area and the third area can be input into the feature extraction network of the target detection network to perform feature extraction to obtain the first point cloud feature of the point cloud data.
  • the feature extraction network includes, for example, a plurality of convolutional layers, and the present disclosure does not limit the structure of the feature extraction network.
  • the point cloud data of the first area and the third area may also be sampled, so as to reduce the amount of processed data.
  • point cloud data with N points is sampled as point cloud data with N/4 points by random sampling.
  • the first point cloud feature may be input into the region generation network RPN of the target detection network for processing, so as to obtain the second candidate frame of the target in the point cloud data of the t-th frame.
  • the number of the second candidate frames is relatively large, which can be further processed.
  • a preset number of first candidate frames may be determined from the second candidate frame, for example, in a non-maximum suppression (Non-maximum suppression, NMS) manner.
  • the preset number may be, for example, 50, which is not limited in the present disclosure.
  • the candidate frame corresponding to the target can be preliminarily estimated from the point cloud data for subsequent processing.
  • step S11 may include:
  • a preset number of first candidate frames are determined from the fourth candidate frames.
  • the object detection can be directly performed on the point cloud data of the t-th frame.
  • the feature extraction network includes, for example, a plurality of convolutional layers, and the present disclosure does not limit the structure of the feature extraction network.
  • the point cloud data of the t-th frame may also be sampled to reduce the amount of processed data.
  • point cloud data with M points is sampled as point cloud data with M/4 points by random sampling.
  • the second point cloud feature may be input into the region generation network RPN of the target detection network for processing, to obtain the fourth candidate frame of the target in the point cloud data of the t-th frame.
  • the number of the fourth candidate frame is relatively large, which can be further processed.
  • a preset number of first candidate frames may be determined from the fourth candidate frame, for example, in a non-maximum suppression (Non-maximum suppression, NMS) manner.
  • the preset number may be, for example, 50, which is not limited in the present disclosure.
  • the candidate frame corresponding to the target can be preliminarily estimated from the point cloud data for subsequent processing.
  • step S12 may include:
  • the prediction candidate frames of each target in the point cloud data of the t-th frame are respectively extended to determine the third candidate frame of each target;
  • the third candidate frame and the second area point cloud corresponding to the area where the third candidate frame is located data and perform candidate frame fusion for each target in the t-th frame of point cloud data to obtain a first detection frame of each target in the t-th frame of point cloud data.
  • a prediction candidate frame is predicted for each target in the first area of the point cloud data of the t-th frame, that is, each target in the first area corresponds to a prediction candidate frame. Predict candidate boxes.
  • the prediction candidate frames of each target may be expanded respectively, so as to increase the number of candidate frames.
  • the pose and scale of the target can be determined according to the predicted candidate frame of the target in the point cloud data of the t-th frame; according to the probability distribution of the pose and scale of the target, a certain variance and The mean value is sampled and expanded to obtain multiple third candidate boxes of the target. In this way, the influence of the error of the predicted candidate frame on the subsequent processing can be reduced, and the probability of matching with the first candidate frame can be increased, thereby improving the stability of the detection result and improving the detection accuracy.
  • the third candidate frame and the first candidate frame may be matched respectively to determine the target corresponding to each first candidate frame.
  • this step may include:
  • the third candidate frame whose intersection ratio with the first candidate frame is greater than or equal to the intersection ratio threshold is determined as the third candidate frame that matches the first candidate frame;
  • the target corresponding to the third candidate frame matching the first candidate frame is determined as the target corresponding to the first candidate frame.
  • the third candidate frame can be matched with the first candidate frame by intersecting and comparing.
  • the intersection-over-union (IoU) ratio between each third candidate frame and each first candidate frame may be determined respectively.
  • a threshold value for example, 0.5
  • the third candidate frame is determined as a candidate frame matching the first candidate frame; and the target corresponding to the third candidate frame is determined as the target corresponding to the first candidate frame.
  • the identification (Identity Document, ID) of the target corresponding to the third candidate frame is assigned to the first candidate frame, that is, it is considered that the two matching candidate frames correspond to the same target.
  • any first candidate frame if there is no third candidate frame whose intersection ratio with the first candidate frame is greater than or equal to the intersection ratio threshold, it can be considered that it is the same as the first candidate frame.
  • the target corresponding to a candidate frame is a new target that has not appeared before. In this case, a new ID can be assigned to the object corresponding to the first candidate frame.
  • the identification of the corresponding target of each first candidate frame can be determined, so as to fuse the candidate frames of the target with the same identification.
  • candidate frame fusion is performed on each target in the t-th frame of point cloud data to obtain the first detection frame of each target in the t-th frame of point cloud data.
  • the target for any target in the point cloud data of the t-th frame, if the target has a first candidate frame and a third candidate frame, it can be segmented from the point cloud data of the t-th frame.
  • the point cloud data of the first area corresponding to the area where the first candidate frame of the target is located, and the point cloud data of the second area corresponding to the area where the third candidate frame of the target is located are segmented.
  • the first detection frame includes a three-dimensional area frame.
  • the first candidate frame related to the target can be segmented from the point cloud data of the t-th frame.
  • the first candidate frame of the target and the point cloud data of the first area are input into the pre-trained fusion network for processing, and the first detection frame of the target is output.
  • the above processing is performed on all the objects in the point cloud data of the t-th frame, and the first detection frame of all the objects in the point-cloud data of the t-th frame can be obtained.
  • the first detection frame of all targets in the point cloud data of the t-th frame may be used as the detection result of the point cloud data of the t-th frame (which may be referred to as the first detection result); other processing may also be performed. (such as classifying objects), so that the detection results of the point cloud data of the t-th frame include more content.
  • This disclosure does not limit this.
  • the first detection frame of all objects in the point cloud data of the t-th frame can be determined, so as to realize the accurate detection of the objects in the point cloud data of the t-th frame.
  • the first detection result further includes the category of the target in the point cloud data of the t-th frame
  • step S12 includes:
  • the objects in the point cloud data of the t-th frame can be classified in step S12.
  • the point cloud data of the t-th frame (which can be called the second target)
  • the point cloud data of the t-th frame can be divided into the area where the first detection frame is located The corresponding third region point cloud data.
  • the point cloud data of the third region may be input into a pre-trained classification network for processing to determine the category to which the second target belongs.
  • the classification network may include, for example, a convolutional layer, a fully connected layer, etc.
  • the present disclosure does not limit the specific network structure of the classification network.
  • the above processing is performed on all the targets in the point cloud data of the t-th frame, and the categories of all the targets in the point-cloud data of the t-th frame can be obtained, so that the categories of the targets are added to the point-cloud data of the t-th frame. the first test result.
  • the detected target information is more abundant.
  • the first detection result can also be combined with the previous historical detection results to further optimize the detection result of the point cloud data of the t-th frame.
  • the target detection method according to the embodiment of the present disclosure may further include:
  • the second detection result is to modify the first detection result of the point cloud data of the t-th frame to determine the second detection result of the point cloud data of the t-th frame.
  • the point cloud data of the first t-1 frames have obtained the final detection result (which can be called the second detection result) in the previous processing, and each second detection result includes the second detection frame of the target, and the t-th frame
  • the target in the point cloud data may have a corresponding second detection frame in the second detection result of the point cloud data of frame t-1.
  • any target in the point cloud data of the t-th frame if there is a second detection frame of the target in the second detection result of the point cloud data of the previous t-1 frame, it can be determined according to The target is the second detection frame in the point cloud data of the previous t-1 frame, and the first detection frame of the target in the point cloud data of the t frame is corrected to obtain the revised detection frame, which is called the second detection frame .
  • the target can be located in the t-th frame of the point cloud data of the t-th frame.
  • a detection frame is directly used as the second detection frame.
  • the above processing is performed on all the targets in the point cloud data of the t-th frame, and the second detection frame of all the targets in the point-cloud data of the t-th frame can be obtained, thereby obtaining the point cloud data of the t-th frame.
  • the second test result is performed on all the targets in the point cloud data of the t-th frame, and the second detection frame of all the targets in the point-cloud data of the t-th frame can be obtained, thereby obtaining the point cloud data of the t-th frame.
  • the first detection result of the t-th frame of point cloud data is based on the second detection result of the t-1 frame of point cloud data before the t-th frame of point cloud data
  • the step of performing correction to determine the second detection result of the point cloud data of the t-th frame may include:
  • the first target is any target in the point cloud data of the t-th frame
  • the detection frame set of the first target includes the first target at the t-1 frame point
  • a detection frame whose error between the detection frame set and the detection frame is less than or equal to the error threshold is determined as the inner point of the detection frame frame;
  • the third detection frame and all the inner point frames of the third detection frame are fused to determine the second detection frame of the first target in the t-th frame of point cloud data.
  • a detection frame set of the first target can be obtained.
  • the detection frame set includes the second detection frame of the first target in the second detection result of the point cloud data of frame t-1, and the first detection frame of the first target in the first detection result of the point cloud data of frame t frame.
  • errors between other detection frames in the detection frame set and the detection frame may be determined.
  • An error threshold can be preset, and a detection frame whose error with the detection frame is less than or equal to the error threshold can be determined as the inner point frame of the detection frame; otherwise, the error with the detection frame is greater than the error threshold.
  • the detection frame of can be determined as the outer point frame of the detection frame. The present disclosure does not limit the specific value of the error threshold.
  • the third detection frame with the largest number of in-point frames may be determined from the detection frame set of the first target, and the third detection frame is used as the initial estimated detection frame.
  • the optimal estimation of the position information of the first target can be obtained, that is, the revised second detection frame can be obtained.
  • the fusion optimization of the third detection frame and all the inner point frames of the third detection frame may be performed by means of least squares, and the third detection frame and the third detection frame may also be optimized by Kalman filtering. Fusion optimization is performed on all the inner point frames of the third detection frame, and the present disclosure does not limit the specific manner of fusion optimization.
  • the above processing is performed on all the targets in the point cloud data of the t-th frame, and the second detection frame of all the targets in the point-cloud data of the t-th frame can be obtained, thereby obtaining the point cloud data of the t-th frame.
  • the second test result is performed on all the targets in the point cloud data of the t-th frame, and the second detection frame of all the targets in the point-cloud data of the t-th frame can be obtained, thereby obtaining the point cloud data of the t-th frame.
  • the detection results can be combined with the previous historical detection results to further optimize the detection results of the point cloud data of the t-th frame and improve the accuracy of target detection.
  • the method further includes:
  • the second detection result of the point cloud data of the t-1 frame and the second detection result of the point cloud data of the t frame predict the motion state of the target in the point cloud data of the t+1 frame, and determine the The predicted candidate frame of the target in the point cloud data of the t+1th frame.
  • the point cloud data of the t+1-th frame can be predicted according to the historical detection results, so as to help the target detection of the point-cloud data of the t+1-th frame.
  • the third target in the second detection result of the point cloud data of the t frame can be obtained.
  • the prediction of the motion state may be implemented by means of Kalman filtering or least squares, which is not limited in the present disclosure.
  • the third target can be The target is predicted, and the third target is predicted to be obtained by the error between the second detection frame in the point cloud data of the t frame and the prediction candidate frame in the point cloud data of the t+1 frame. +1 frame of predicted candidate boxes in point cloud data.
  • the prediction candidate frames of all the objects in the detected area of the point cloud data of the t+1-th frame can be determined.
  • the prediction candidate frame of the target in the point cloud data of the t+1th frame can be obtained, which can help the target detection of the point cloud data of the t+1th frame, thereby improving the detection accuracy.
  • the method further includes:
  • the prediction candidate frame of the target in the t+1th frame of point cloud data and the tth frame of point cloud data update the prediction probability map of the target in the tth frame of point cloud data, and determine the t+1th frame Predicted probability map of objects in frame point cloud data.
  • the prediction probability map of the target in the point cloud data of the t frame can be performed according to the prediction candidate frame and the point cloud data of the t frame. renew. That is, according to the position of the target in the point cloud data of the t frame and the position of the point cloud data of the t+1 frame (prediction candidate frame), determine whether there is a target at each position in the prediction probability map, and update each position may appear.
  • the probability of the target is obtained to obtain the predicted probability map of the target in the point cloud data of the t+1th frame.
  • the predicted probability map of the target in the t+1th frame of point cloud data can be obtained, so as to divide multiple regions for the t+1th frame of point cloud data in subsequent processing, thereby improving the speed of target detection.
  • the method further includes:
  • the motion state of the target in the t-th frame of point cloud data is predicted, and the t-th frame of point cloud data is determined.
  • Predicted candidate boxes for objects in the data are
  • the point cloud data of the t frame can be predicted according to the historical detection results, so as to help the target detection of the point cloud data of the t frame.
  • the second detection frame of the target in the second detection result of the point cloud data of the previous t-1 frame can be obtained.
  • the motion state is predicted, the position of the target in the point cloud data of the t frame is predicted, and the prediction candidate frame of the target in the point cloud data of the t frame is obtained.
  • the prediction process is similar to the prediction process for the point cloud data of the t+1th frame, and the description is not repeated here.
  • the predicted candidate frame of the target in the point cloud data of the t frame can be obtained, which can help the target detection of the point cloud data of the t frame, thereby improving the detection accuracy.
  • the method further includes:
  • the predicted candidate frame of the target in the t-th frame of point cloud data and the t-1-th frame of point cloud data update the predicted probability map of the target in the t-1-th frame of point cloud data, and determine the t-th frame Predicted probability map of objects in point cloud data.
  • the predicted probability of the target in the t-1 th frame of point cloud data can be calculated according to the predicted candidate frame and the t-1 th frame of point cloud data.
  • the graph is updated to obtain the predicted probability graph of the target in the point cloud data of the t-th frame.
  • the update process is similar to the update process of the prediction probability map of the point cloud data of the t+1th frame, and the description is not repeated here.
  • the predicted probability map of the target in the t-th frame of point cloud data can be obtained, so as to divide multiple regions for the t-th frame of point cloud data in subsequent processing, thereby improving the speed of target detection.
  • FIG. 1B shows a schematic diagram of a network architecture of the target detection method according to the embodiment of the present disclosure, and the network architecture includes: User terminal 201 , network 202 and target detection terminal 203 .
  • the target detection terminal 203 obtains the first candidate frame by predicting the detection frame of the target; finally, the target detection terminal 203 passes the first candidate frame.
  • the candidate frame is used to detect the target in the point cloud data of the t-th frame, and obtain the detection result of the target. In this way, the detection frame is predicted by using multi-frame point cloud data, which can improve the accuracy of target detection.
  • FIG. 2 shows a schematic diagram of a processing procedure of a target detection method according to an embodiment of the present disclosure.
  • the process of performing target detection processing on the current frame is called the front end; the process of recording the historical results, revising the current frame and predicting the next frame according to the historical results is called the back end,
  • the back-end processing can also be referred to as object tracking and fusion.
  • the current frame is the t-th frame.
  • the first detection result (not shown) of the point cloud data of the t-1th frame is obtained in the front-end processing of the t-1th frame before; the first detection result is compared with the history of the previous t-2 frame.
  • the detection results are correlated, and in step 211 at the back end of the t-1 frame, the fusion optimization of the detection frame is performed by Kalman filtering or least squares, so as to realize the correction of the detection results, and obtain the second point cloud data of the t-1 frame. Test results (not shown).
  • the motion prediction 212 of the target in the t-th frame can be performed according to the historical detection results of the previous t-1 frame, and the motion prediction of the target in the t-th frame of point cloud data can be obtained.
  • the predicted probability map 215 of the target in the middle thus completing the entire processing process of the t-1th frame.
  • the point cloud data 221 of the t-th frame may be divided into a first area where the target exists, a second area where the target does not exist, and whether there is an undetermined presence or not according to the prediction probability map 215 .
  • the divided area point cloud data 222 is obtained.
  • the first area and the third area of the point cloud data 222 are input into the target detection network 223 for target detection, and a preset number of first candidate frames can be obtained.
  • the first detection result 226 may be associated with the historical detection result of the previous t-1 frames.
  • the fusion and optimization of the detection frame can be performed by means of Kalman filtering or least squares, so as to realize the correction of the detection result, and obtain the point cloud data of the t-th frame.
  • the second detection frame of each target in is taken as the second detection result 230 of the point cloud data of the t-th frame, that is, the final output result.
  • motion prediction 232 can be performed on the target in the t+1-th frame according to the second detection result of the previous t-frame to obtain the target in the t+1-th frame point cloud data Then according to the prediction candidate frame 233 and the t-th frame point cloud data 221, in step 234, the prediction probability map 215 of the t-th frame is updated to obtain the prediction of the target in the t+1-th frame point cloud data probability map 235, thus completing the entire processing of the t-th frame.
  • Figure 3a shows a schematic diagram of an image of a target scene
  • Figure 3b shows a schematic diagram of a detection result of the target.
  • the target scene includes multiple chairs, and the chairs can be used as the target to be detected.
  • the detection frame 31 is the detection result obtained by the target detection method of single frame processing according to the related art
  • the detection frame 32 is the real three-dimensional image frame of the target
  • the detection frame 33 is the target detection method according to the embodiment of the present disclosure obtained test results.
  • the detection result obtained by the target detection method of the embodiment of the present disclosure has higher precision.
  • the detection result of the related art is obviously deteriorated, but the target detection method of the embodiment of the present disclosure can still maintain high accuracy.
  • the historical detection results can be effectively used to detect and track the 3D target; the historical detection results can be used to predict and track the 3D target.
  • the candidate frame of the target in the current frame, and the distribution map of the probability that 3D objects may appear in the known area in the current frame, are fed back to the target detection process of the current frame; it enables the current frame to use the predicted probability distribution during target detection.
  • the image is divided into regions, thereby reducing the amount of data processed and improving the speed of target detection; and using the predicted candidate frame as the a priori frame, not only avoids the target search of the entire scene for each frame, but also obtains more accurate candidates according to the a priori frame. frame, which effectively improves the accuracy of target detection and avoids the occurrence of missed detection.
  • the tracking and fusion of the target can be performed, and all the detection frames of each 3D target in continuous time are stored as the historical detection frames of the 3D object, and the All historical detection frames of each 3D target are fused and optimized to obtain the optimal estimate of the position of the 3D target in the current frame, thereby effectively improving the stability of the 3D detection frame and reducing the detection error when the target is occluded or truncated.
  • the target detection method according to the embodiment of the present disclosure can be applied to application scenarios such as augmented reality AR, indoor navigation, etc., to realize 3D target estimation and detection.
  • the processing method of the related art does not consider the relationship between the position information of the same object in consecutive frames, and does not utilize the information on the continuous time, which easily causes the jitter of the 3D detection frame. For example, in indoor scenes, due to the larger size of the object, the phenomenon of detection frame jitter will be more serious.
  • a more stable 3D detection frame can be output by utilizing the relationship of position information in consecutive frames and the information in continuous time, and the detection error can be reduced.
  • the present disclosure also provides target detection devices, electronic devices, computer-readable storage media, and programs, all of which can be used to implement any target detection method provided by the present disclosure.
  • target detection devices electronic devices, computer-readable storage media, and programs, all of which can be used to implement any target detection method provided by the present disclosure.
  • FIG. 4 shows a block diagram of a target detection apparatus according to an embodiment of the present disclosure. As shown in FIG. 4 , the apparatus includes:
  • the first detection module 41 is configured to perform target detection on the t-th frame of point cloud data of the target scene, and determine the first candidate frame of the target in the t-th frame of point cloud data, where t is an integer greater than 1;
  • the second detection module 42 is configured to determine the point cloud data of the t frame according to the point cloud data of the t frame, the first candidate frame and the predicted candidate frame for the target in the point cloud data of the t frame.
  • the first detection result includes the first detection frame of the target in the point cloud data of the t-th frame, wherein the prediction candidate frame is based on the t-th frame before the point cloud data of the t-th frame.
  • the detection results of 1 frame of point cloud data are predicted.
  • the first detection module includes: a region division sub-module, configured to divide the point cloud data of the t-th frame according to the predicted probability map of the target in the point-cloud data of the t-th frame is a first area with a target, a second area without a target, and a third area where it is not determined whether there is a target; a first detection submodule is configured to perform target detection on the first area and the third area, Determine the first candidate frame of the target in the point cloud data of the t-th frame.
  • the apparatus further includes: a correction module configured to acquire the second detection result of the point cloud data of the t-1 frame before the point cloud data of the t-th frame; The second detection result of the t-1 frame of point cloud data before the t-th frame of point cloud data, the first detection result of the t-th frame of point cloud data is corrected, and the second detection result of the t-th frame of point cloud data is determined. Test results.
  • the apparatus further includes: a first motion prediction module configured to, according to the second detection result of the point cloud data of the t-1 frame before the point cloud data of the t frame, perform a The motion state of the target in the point cloud data of the t-th frame is predicted, and the prediction candidate frame of the target in the point cloud data of the t-th frame is determined.
  • a first motion prediction module configured to, according to the second detection result of the point cloud data of the t-1 frame before the point cloud data of the t frame, perform a The motion state of the target in the point cloud data of the t-th frame is predicted, and the prediction candidate frame of the target in the point cloud data of the t-th frame is determined.
  • the apparatus further includes: a first probability map update module, configured to predict candidate frames of the target in the t-th frame of point cloud data, and the t-1-th frame of point cloud data, The predicted probability map of the target in the t-1th frame of point cloud data is updated to determine the predicted probability map of the target in the t-th frame of point cloud data.
  • a first probability map update module configured to predict candidate frames of the target in the t-th frame of point cloud data, and the t-1-th frame of point cloud data, The predicted probability map of the target in the t-1th frame of point cloud data is updated to determine the predicted probability map of the target in the t-th frame of point cloud data.
  • the first detection sub-module is configured to: perform feature extraction on the point cloud data of the first area and the third area to obtain a first point cloud feature;
  • the first point cloud feature is used for target detection, and the second candidate frame of the target in the point cloud data of the t-th frame is determined; according to the confidence of each second candidate frame, a preset number of The first candidate box.
  • the second detection module includes: a candidate frame expansion sub-module, configured to expand the prediction candidate frames of each target in the point cloud data of the t-th frame respectively, and determine the th Three candidate frames; a candidate frame matching submodule, configured to match the third candidate frame and the first candidate frame respectively, to determine the target corresponding to each first candidate frame; the candidate frame fusion submodule, configured to The first candidate frame and the first region point cloud data corresponding to the region where the first candidate box is located, and the third candidate box and the second region point cloud data corresponding to the region where the third candidate box is located , performing candidate frame fusion on each target in the t-th frame of point cloud data, to obtain a first detection frame of each target in the t-th frame of point cloud data.
  • the candidate frame matching sub-module is configured to: respectively determine the intersection ratio between each third candidate frame and each first candidate frame; compare the intersection ratio with the first candidate frame The third candidate frame that is greater than or equal to the intersection ratio threshold is determined as the third candidate frame that matches the first candidate frame; the target corresponding to the third candidate frame that matches the first candidate frame is determined to match the The target corresponding to the first candidate frame.
  • each second detection result includes a second detection frame of the target
  • the correction module includes: a set determination submodule configured to determine a detection frame set of a first target, the first target is any target in the t-th frame of point cloud data, and the detection frame set of the first target includes the second detection frame of the first target in the second detection result of the t-1 frame of point cloud data, and the first detection frame of the first target in the first detection result of the point cloud data of the t-th frame;
  • the inner point frame determination submodule is configured to be any detection frame in the detection frame set for the first target, A detection frame whose error between the detection frame set and the detection frame is less than or equal to an error threshold is determined as the inner point frame of the detection frame;
  • the detection frame selection sub-module is configured to select the detection frame from the first target Determine the third detection frame with the largest number of in-point frames in the detection frame set of the The second detection frame of the first target in the point cloud data of the t-th frame.
  • the apparatus further includes: a second motion prediction module, configured to be based on the second detection result of the point cloud data of the t-1 frame, and the second detection result of the point cloud data of the t frame The second detection result is to predict the motion state of the target in the t+1th frame of point cloud data, and determine the prediction candidate frame of the target in the t+1th frame of point cloud data.
  • a second motion prediction module configured to be based on the second detection result of the point cloud data of the t-1 frame, and the second detection result of the point cloud data of the t frame The second detection result is to predict the motion state of the target in the t+1th frame of point cloud data, and determine the prediction candidate frame of the target in the t+1th frame of point cloud data.
  • the apparatus further includes: a second probability map update module, configured to predict candidate frames of the target in the t+1 th frame of point cloud data, and the t th frame of point cloud data, The predicted probability map of the target in the t-th frame of point cloud data is updated, and the predicted probability map of the target in the t+1-th frame of point cloud data is determined.
  • a second probability map update module configured to predict candidate frames of the target in the t+1 th frame of point cloud data, and the t th frame of point cloud data, The predicted probability map of the target in the t-th frame of point cloud data is updated, and the predicted probability map of the target in the t+1-th frame of point cloud data is determined.
  • the first detection module includes: a feature extraction sub-module configured to perform feature extraction on the point cloud data of the t-th frame to obtain second point cloud features; the second detection sub-module, is configured to perform target detection on the second point cloud feature, and determine the fourth candidate frame of the target in the point cloud data of the t-th frame; the selection sub-module is configured to, according to the confidence of each fourth candidate frame, select the target frame from the The fourth candidate frame determines a preset number of first candidate frames.
  • the first detection result further includes the category of the target in the point cloud data of the t-th frame
  • the second detection module includes: a classification sub-module configured to The third area point cloud data corresponding to the area where the first detection frame is located, classify the second target, determine the category of the second target, and the second target is any point cloud data in the t-th frame. a goal.
  • the target scene includes an indoor scene
  • the target in the t-th frame of point cloud data includes an object
  • the first detection frame of the target in the t-th frame of point cloud data includes a three-dimensional area frame .
  • the functions or modules included in the apparatuses provided in the embodiments of the present disclosure may be used to execute the methods described in the above method embodiments.
  • Embodiments of the present disclosure further provide a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the foregoing method is implemented.
  • the computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • An embodiment of the present disclosure further provides an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to invoke the instructions stored in the memory to execute the above method.
  • Embodiments of the present disclosure also provide a computer program product, including computer-readable codes.
  • a processor in the device executes a method for implementing the target detection method provided in any of the above embodiments. instruction.
  • Embodiments of the present disclosure further provide another computer program product for storing computer-readable instructions, which, when executed, cause the computer to perform the operations of the target detection method provided by any of the foregoing embodiments.
  • the electronic device may be provided as a terminal, server or other form of device.
  • FIG. 5 shows a block diagram of an electronic device 800 according to an embodiment of the present disclosure.
  • electronic device 800 may be a mobile phone, computer, digital broadcast terminal, messaging device, game console, tablet device, medical device, fitness device, personal digital assistant, etc. terminal.
  • electronic device 800 may include one or more of the following components: processing component 802, memory 804, power supply component 806, multimedia component 808, audio component 810, input/output (I/O) interface 812, sensor component 814 , and the communication component 816 .
  • the processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations.
  • the processing component 802 can include one or more processors 820 to execute instructions to perform all or some of the steps of the methods described above.
  • processing component 802 may include one or more modules that facilitate interaction between processing component 802 and other components.
  • processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.
  • Memory 804 is configured to store various types of data to support operation at electronic device 800 . Examples of such data include instructions for any application or method operating on electronic device 800, contact data, phonebook data, messages, pictures, videos, and the like. Memory 804 may be implemented by any type of volatile or nonvolatile storage device or combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Magnetic Memory, Flash Memory, Magnetic or Optical Disk.
  • SRAM static random access memory
  • EEPROM electrically erasable programmable read only memory
  • EPROM erasable Programmable Read Only Memory
  • PROM Programmable Read Only Memory
  • ROM Read Only Memory
  • Magnetic Memory Flash Memory
  • Magnetic or Optical Disk Magnetic Disk
  • Power supply assembly 806 provides power to various components of electronic device 800 .
  • Power supply components 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 800 .
  • Multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user.
  • the touch panel includes one or more touch sensors to sense touch, swipe, and gestures on the touch panel. The touch sensor may not only sense the boundaries of a touch or swipe action, but also detect the duration and pressure associated with the touch or swipe action.
  • the multimedia component 808 includes a front-facing camera and/or a rear-facing camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera may receive external multimedia data. Each of the front and rear cameras can be a fixed optical lens system or have focal length and optical zoom capability.
  • Audio component 810 is configured to output and/or input audio signals.
  • audio component 810 includes a microphone (MIC) that is configured to receive external audio signals when electronic device 800 is in operating modes, such as calling mode, recording mode, and voice recognition mode.
  • the received audio signal may be further stored in memory 804 or transmitted via communication component 816 .
  • audio component 810 also includes a speaker for outputting audio signals.
  • the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module, which may be a keyboard, a click wheel, a button, or the like. These buttons may include, but are not limited to: home button, volume buttons, start button, and lock button.
  • Sensor assembly 814 includes one or more sensors for providing status assessment of various aspects of electronic device 800 .
  • the sensor assembly 814 can detect the on/off state of the electronic device 800, the relative positioning of the components, such as the display and the keypad of the electronic device 800, the sensor assembly 814 can also detect the electronic device 800 or one of the electronic device 800 Changes in the position of components, presence or absence of user contact with the electronic device 800 , orientation or acceleration/deceleration of the electronic device 800 and changes in the temperature of the electronic device 800 .
  • Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects in the absence of any physical contact.
  • Sensor assembly 814 may also include a light sensor, such as a complementary metal oxide semiconductor (CMOS) or charge coupled device (CCD) image sensor, for use in imaging applications.
  • CMOS complementary metal oxide semiconductor
  • CCD charge coupled device
  • the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • Communication component 816 is configured to facilitate wired or wireless communication between electronic device 800 and other devices.
  • the electronic device 800 may access a wireless network based on a communication standard, such as wireless network (WiFi), second generation mobile communication technology (2G) or third generation mobile communication technology (3G), or a combination thereof.
  • the communication component 816 receives broadcast signals or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication.
  • the NFC module may be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
  • RFID radio frequency identification
  • IrDA infrared data association
  • UWB ultra-wideband
  • Bluetooth Bluetooth
  • electronic device 800 may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable A programmed gate array (FPGA), controller, microcontroller, microprocessor or other electronic component implementation is used to perform the above method.
  • ASICs application specific integrated circuits
  • DSPs digital signal processors
  • DSPDs digital signal processing devices
  • PLDs programmable logic devices
  • FPGA field programmable A programmed gate array
  • controller microcontroller, microprocessor or other electronic component implementation is used to perform the above method.
  • a non-volatile computer-readable storage medium such as a memory 804 comprising computer program instructions executable by the processor 820 of the electronic device 800 to perform the above method is also provided.
  • FIG. 6 shows a block diagram of an electronic device 1900 according to an embodiment of the present disclosure.
  • the electronic device 1900 may be provided as a server.
  • electronic device 1900 includes processing component 1922, which further includes one or more processors, and a memory resource represented by memory 1932 for storing instructions executable by processing component 1922, such as applications.
  • An application program stored in memory 1932 may include one or more modules, each corresponding to a set of instructions.
  • the processing component 1922 is configured to execute instructions to perform the above-described methods.
  • the electronic device 1900 may also include a power supply assembly 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input output (I/O) interface 1958 .
  • the electronic device 1900 can operate based on an operating system stored in the memory 1932, such as a Microsoft server operating system (Windows ServerTM), a graphical user interface based operating system (Mac OS XTM) introduced by Apple, a multi-user multi-process computer operating system (UnixTM). ), Free and Open Source Unix-like Operating System (LinuxTM), Open Source Unix-like Operating System (FreeBSDTM) or similar.
  • a non-volatile computer-readable storage medium such as memory 1932 comprising computer program instructions executable by processing component 1922 of electronic device 1900 to perform the above-described method.
  • the present disclosure may be a system, method and/or computer program product.
  • the computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the present disclosure.
  • a computer-readable storage medium may be a tangible device that can hold and store instructions for use by the instruction execution device.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • Non-exhaustive list of computer readable storage media include: portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM) or flash memory), static random access memory (SRAM), portable compact disk read only memory (CD-ROM), digital versatile disk (DVD), memory sticks, floppy disks, mechanically coded devices, such as printers with instructions stored thereon Hole cards or raised structures in grooves, and any suitable combination of the above.
  • RAM random access memory
  • ROM read only memory
  • EPROM erasable programmable read only memory
  • flash memory static random access memory
  • SRAM static random access memory
  • CD-ROM compact disk read only memory
  • DVD digital versatile disk
  • memory sticks floppy disks
  • mechanically coded devices such as printers with instructions stored thereon Hole cards or raised structures in grooves, and any suitable combination of the above.
  • Computer-readable storage media are not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (eg, light pulses through fiber optic cables), or through electrical wires transmitted electrical signals.
  • the computer readable program instructions described herein may be downloaded to various computing/processing devices from a computer readable storage medium, or to an external computer or external storage device over a network such as the Internet, a local area network, a wide area network, and/or a wireless network.
  • the network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers.
  • a network adapter card or network interface in each computing/processing device receives computer-readable program instructions from a network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing/processing device .
  • Computer program instructions for carrying out operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or instructions in one or more programming languages.
  • Source or object code written in any combination, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as the "C" language or similar programming languages.
  • the computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server implement.
  • the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (eg, using an Internet service provider through the Internet connect).
  • LAN local area network
  • WAN wide area network
  • custom electronic circuits such as programmable logic circuits, field programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) can be personalized by utilizing state information of computer readable program instructions.
  • Computer readable program instructions are executed to implement various aspects of the present disclosure.
  • These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer or other programmable data processing apparatus to produce a machine that causes the instructions when executed by the processor of the computer or other programmable data processing apparatus , resulting in means for implementing the functions/acts specified in one or more blocks of the flowchart and/or block diagrams.
  • These computer readable program instructions can also be stored in a computer readable storage medium, these instructions cause a computer, programmable data processing apparatus and/or other equipment to operate in a specific manner, so that the computer readable medium storing the instructions includes An article of manufacture comprising instructions for implementing various aspects of the functions/acts specified in one or more blocks of the flowchart and/or block diagrams.
  • Computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other equipment to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other equipment to produce a computer-implemented process , thereby causing instructions executing on a computer, other programmable data processing apparatus, or other device to implement the functions/acts specified in one or more blocks of the flowcharts and/or block diagrams.
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more functions for implementing the specified logical function(s) executable instructions.
  • the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.
  • each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations can be implemented in dedicated hardware-based systems that perform the specified functions or actions , or can be implemented in a combination of dedicated hardware and computer instructions.
  • the computer program product can be specifically implemented by hardware, software or a combination thereof.
  • the computer program product is embodied as a computer storage medium, and in another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (Software Development Kit, SDK), etc. Wait.
  • a software development kit Software Development Kit, SDK
  • the present disclosure relates to a target detection method and device, electronic equipment and storage medium.
  • the method includes: performing target detection on point cloud data of the t-th frame of a target scene, and determining a first target of the target in the point-cloud data of the t-th frame.
  • candidate frame, t is an integer greater than 1; according to the point cloud data of the t frame, the first candidate frame and the predicted candidate frame for the target in the point cloud data of the t frame, determine the point of the t frame
  • the first detection result of the cloud data, the first detection result includes the first detection frame of the target in the point cloud data of the t-th frame, wherein the prediction candidate frame is based on the point cloud data of the t-th frame before.
  • the detection result of the point cloud data of frame t-1 is predicted.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Image Analysis (AREA)

Abstract

本公开涉及一种目标检测方法及装置、电子设备和存储介质，所述方法包括：对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，t为大于1的整数；根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框，其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的；如此，通过多帧点云数据进行检测框的预测，能够提高目标检测的精度。

Description

目标检测方法及装置、电子设备和存储介质
相关申请的交叉引用
本公开基于申请号为202010738105.2、申请日为2020年7月28日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本公开作为参考。
技术领域
本公开涉及计算机技术领域，尤其涉及一种目标检测方法及装置、电子设备和存储介质。
背景技术
目标检测是计算机视觉中的一个非常重要的任务，能够通过传感器的输入数据，对视野范围内的目标(例如人或物体)的位姿、尺度等信息进行估计。在相关技术中，目标检测方法通常将每一帧的输入单独进行处理，导致检测精度较差。
发明内容
本公开提出了一种目标检测技术方案。
根据本公开的一方面，提供了一种目标检测方法，包括：对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，t为大于1的整数；根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框，其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的。如此，通过多帧点云数据进行检测框的预测，能够提高目标检测的精度。
在一种可能的实现方式中，所述对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，包括：根据所述第t帧点云数据中目标的预测概率图，将所述第t帧点云数据划分为存在目标的第一区域、不存在目标的第二区域以及未确定是否存在目标的第三区域；对所述第一区域及所述第三区域进行目标检测，确定所述第t帧点云数据中目标的第一候选框。如此，可以减少目标检测所处理的点云数据的数据量，能够提高检测速度。
在一种可能的实现方式中，所述方法还包括：获取在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果；根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果。如果，通过进一步修正第一检测结果，能够进一步提高第一检测结果的准确度。
在一种可能的实现方式中，所述方法还包括：根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据中目标的运动状态进行预测，确定所述第t帧点云数据中目标的预测候选框。如此，基于前面多帧点云数据预测第t帧的点云数据中目标的预测候选框，能够提高预测的准确度。
在一种可能的实现方式中，所述方法还包括：根据所述第t帧点云数据中目标的预测候选框，以及第t-1帧点云数据，对所述第t-1帧点云数据中目标的预测概率图进行更新，确定所述第t帧点云数据中目标的预测概率图。如此，基于前面的多帧点云数据对预测概率图进行更新，能够进一步预测目标出现的概率，以提高最终得到的预测概率图的准确度。
在一种可能的实现方式中，所述对所述第一区域及所述第三区域进行目标检测，确定所述第t帧点云数据中目标的第一候选框，包括：对所述第一区域及所述第三区域的点 云数据进行特征提取，得到第一点云特征；对所述第一点云特征进行目标检测，确定所述第t帧点云数据中目标的第二候选框；根据各个第二候选框的置信度，从所述第二候选框中确定出预设数量的第一候选框。如此，将采样后的点云数据输入特征提取网络中处理，得到第一点云特征；能够进一步提高检测速度。
在一种可能的实现方式中，所述根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，包括：对所述第t帧点云数据中各个目标的预测候选框分别进行扩展，确定各个目标的第三候选框；对所述第三候选框与所述第一候选框分别进行匹配，确定与各个第一候选框对应的目标；根据所述第一候选框及与所述第一候选框所在区域对应的第一区域点云数据，以及所述第三候选框及与所述第三候选框所在区域对应的第二区域点云数据，对所述第t帧点云数据中的各个目标分别进行候选框融合，得到所述第t帧点云数据中各个目标的第一检测框。如此，能够提高与第一候选框匹配的概率，以提高检测结果的稳定性。
在一种可能的实现方式中，所述对所述第三候选框与所述第一候选框分别进行匹配，确定与各个第一候选框对应的目标，包括：分别确定各个第三候选框与各个第一候选框之间的交并比；将与第一候选框的交并比大于或等于交并比阈值的第三候选框，确定为与第一候选框相匹配的第三候选框；将与第一候选框相匹配的第三候选框对应的目标，确定为与所述第一候选框对应的目标。如此，通过将交并比较大的候选框确定为匹配的候选框，从而能够提高预测候选框的准确度。
在一种可能的实现方式中，每个第二检测结果包括目标的第二检测框，所述根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果，包括：确定第一目标的检测框集合，所述第一目标为所述第t帧点云数据中的任意一个目标，所述第一目标的检测框集合包括所述第一目标在t-1帧点云数据的第二检测结果中的第二检测框，以及所述第一目标在第t帧点云数据的第一检测结果中的第一检测框；对于所述第一目标的检测框集合中任意一个检测框，将所述检测框集合中与所述检测框之间的误差小于或等于误差阈值的检测框，确定为所述检测框的内点框；从所述第一目标的检测框集合中确定出内点框数量最多的第三检测框；对所述第三检测框及所述第三检测框的所有内点框进行融合，确定所述第t帧点云数据中第一目标的第二检测框。如此，能够检测出更为丰富的目标信息。
在一种可能的实现方式中，所述方法还包括：根据所述t-1帧点云数据的第二检测结果，以及所述第t帧点云数据的第二检测结果，对第t+1帧点云数据中目标的运动状态进行预测，确定所述第t+1帧点云数据中目标的预测候选框。如此，通过前一帧点云数据的检测结果，可实现对后一帧点云数据中目标的运动状态进行预测，进而实现对后一帧点云数据中目标的检测。
在一种可能的实现方式中，所述方法还包括：根据所述第t+1帧点云数据中目标的预测候选框，以及第t帧点云数据，对所述第t帧点云数据中目标的预测概率图进行更新，确定所述第t+1帧点云数据中目标的预测概率图。如此，通过得到第t+1帧点云数据中目标的预测概率图，以便在后续处理时为第t+1帧点云数据划分多个区域，从而提高目标检测的速度。
在一种可能的实现方式中，所述对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，包括：对所述第t帧点云数据进行特征提取，得到第二点云特征；对所述第二点云特征进行目标检测，确定所述第t帧点云数据中目标的第四候选框；根据各个第四候选框的置信度，从所述第四候选框确定出预设数量的第一候选框。如此，可在未对第t帧点云数据进行区域划分的情况下，实现对第t帧点云数据进行目标检测。
在一种可能的实现方式中，所述第一检测结果还包括所述第t帧点云数据中目标的类别，所述根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，包括：根据与第二目标的第一检测框所在区域对应的第三区域点云数据，对所述第二目标进行分类，确定所述第二目标的类别，所述第二目标为所述第t帧点云数据中的任意一个目标。如此，对于第t帧点云数据中的任意一个目标，能够实现通过该第二目标的第一检测框，从第t帧点云数据中分割出第一检测框所在区域对应的第三区域点云数据；从而能够以较高的准确度实现对第t帧点云数据中任一目标的检测。
在一种可能的实现方式中，所述目标场景包括室内场景，所述第t帧点云数据中的目标包括物体，所述第t帧点云数据中目标的第一检测框包括三维区域框。
根据本公开的一方面，提供了一种目标检测装置，包括：
第一检测模块，配置为对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，t为大于1的整数；
第二检测模块，配置为根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框，
其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的。
在一种可能的实现方式中，所述第一检测模块包括：区域划分子模块，配置为根据所述第t帧点云数据中目标的预测概率图，将所述第t帧点云数据划分为存在目标的第一区域、不存在目标的第二区域以及未确定是否存在目标的第三区域；第一检测子模块，配置为对所述第一区域及所述第三区域进行目标检测，确定所述第t帧点云数据中目标的第一候选框。
在一种可能的实现方式中，所述装置还包括：修正模块，配置为获取在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果；并根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果。
在一种可能的实现方式中，所述装置还包括：第一运动预测模块，配置为根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据中目标的运动状态进行预测，确定所述第t帧点云数据中目标的预测候选框。
在一种可能的实现方式中，所述装置还包括：第一概率图更新模块，配置为根据所述第t帧点云数据中目标的预测候选框，以及第t-1帧点云数据，对所述第t-1帧点云数据中目标的预测概率图进行更新，确定所述第t帧点云数据中目标的预测概率图。
在一种可能的实现方式中，所述第一检测子模块，配置为：对所述第一区域及所述第三区域的点云数据进行特征提取，得到第一点云特征；对所述第一点云特征进行目标检测，确定所述第t帧点云数据中目标的第二候选框；根据各个第二候选框的置信度，从所述第二候选框中确定出预设数量的第一候选框。
在一种可能的实现方式中，所述第二检测模块包括：候选框扩展子模块，配置为对所述第t帧点云数据中各个目标的预测候选框分别进行扩展，确定各个目标的第三候选框；候选框匹配子模块，配置为对所述第三候选框与所述第一候选框分别进行匹配，确定与各个第一候选框对应的目标；候选框融合子模块，配置为根据所述第一候选框及与所述第一候选框所在区域对应的第一区域点云数据，以及所述第三候选框及与所述第三候选框所在区域对应的第二区域点云数据，对所述第t帧点云数据中的各个目标分别进行候选框融合，得到所述第t帧点云数据中各个目标的第一检测框。
在一种可能的实现方式中，所述候选框匹配子模块，配置为：分别确定各个第三候 选框与各个第一候选框之间的交并比；将与第一候选框的交并比大于或等于交并比阈值的第三候选框，确定为与第一候选框相匹配的第三候选框；将与第一候选框相匹配的第三候选框对应的目标，确定为与所述第一候选框对应的目标。
在一种可能的实现方式中，每个第二检测结果包括目标的第二检测框，所述修正模块包括：集合确定子模块，配置为确定第一目标的检测框集合，所述第一目标为所述第t帧点云数据中的任意一个目标，所述第一目标的检测框集合包括所述第一目标在t-1帧点云数据的第二检测结果中的第二检测框，以及所述第一目标在第t帧点云数据的第一检测结果中的第一检测框；内点框确定子模块，配置为对于所述第一目标的检测框集合中任意一个检测框，将所述检测框集合中与所述检测框之间的误差小于或等于误差阈值的检测框，确定为所述检测框的内点框；检测框选择子模块，配置为从所述第一目标的检测框集合中确定出内点框数量最多的第三检测框；内点框融合子模块，配置为对所述第三检测框及所述第三检测框的所有内点框进行融合，确定所述第t帧点云数据中第一目标的第二检测框。
在一种可能的实现方式中，所述装置还包括：第二运动预测模块，配置为根据所述t-1帧点云数据的第二检测结果，以及所述第t帧点云数据的第二检测结果，对第t+1帧点云数据中目标的运动状态进行预测，确定所述第t+1帧点云数据中目标的预测候选框。
在一种可能的实现方式中，所述装置还包括：第二概率图更新模块，配置为根据所述第t+1帧点云数据中目标的预测候选框，以及第t帧点云数据，对所述第t帧点云数据中目标的预测概率图进行更新，确定所述第t+1帧点云数据中目标的预测概率图。
在一种可能的实现方式中，所述第一检测模块包括：特征提取子模块，配置为对所述第t帧点云数据进行特征提取，得到第二点云特征；第二检测子模块，配置为对所述第二点云特征进行目标检测，确定所述第t帧点云数据中目标的第四候选框；选择子模块，配置为根据各个第四候选框的置信度，从所述第四候选框确定出预设数量的第一候选框。
在一种可能的实现方式中，所述第一检测结果还包括所述第t帧点云数据中目标的类别，所述第二检测模块包括：分类子模块，配置为根据与第二目标的第一检测框所在区域对应的第三区域点云数据，对所述第二目标进行分类，确定所述第二目标的类别，所述第二目标为所述第t帧点云数据中的任意一个目标。
在一种可能的实现方式中，所述目标场景包括室内场景，所述第t帧点云数据中的目标包括物体，所述第t帧点云数据中目标的第一检测框包括三维区域框。
根据本公开的一方面，提供了一种电子设备，包括：处理器；配置为存储处理器可执行指令的存储器；其中，所述处理器被配置为调用所述存储器存储的指令，以执行上述方法。
根据本公开的一方面，提供了一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述方法。
根据本公开的实施例，能够检测出第t帧点云数据中目标的第一候选框；通过历史检测结果预测得到的预测候选框对第一候选框进行修正，得到第t帧点云数据的检测结果，从而提高目标检测的精度。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本公开。根据下面参考附图对示例性实施例的详细说明，本公开的其它特征及方面将变得清楚。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。
图1A示出根据本公开实施例的目标检测方法的流程图。
图1B示出本公开实施例目标检测方法的一种网络架构示意图；
图2示出根据本公开实施例的目标检测方法的处理过程的示意图。
图3a示出目标场景的图像的示意图。
图3b示出目标的检测结果的示意图。
图4示出根据本公开实施例的目标检测装置的框图。
图5示出根据本公开实施例的一种电子设备的框图。
图6示出根据本公开实施例的一种电子设备的框图。
具体实施方式
以下将参考附图详细说明本公开的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合，例如，包括A、B、C中的至少一种，可以表示包括从A、B和C构成的集合中选择的任意一个或多个元素。
另外，为了更好地说明本公开，在下文的具体实施方式中给出了众多的具体细节。本领域技术人员应当理解，没有某些具体细节，本公开同样可以实施。在一些实例中，对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述，以便于凸显本公开的主旨。
图1A示出根据本公开实施例的目标检测方法的流程图，如图1A所示，所述目标检测方法包括：
在步骤S11中，对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框；
在本公开实施例中，t为大于1的整数。
在步骤S12中，根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框；
其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的。
在一种可能的实现方式中，所述目标检测方法可以由终端设备或服务器等电子设备执行，终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、蜂窝电话、无绳电话、个人数字处理(Personal Digital Assistant，PDA)、手持设备、计算设备、车载设备、可穿戴设备等，所述方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。或者，可通过服务器执行所述方法。
举例来说，目标场景可包括商场、医院、展馆等室内场景，也可包括交通枢纽、城市街道等室外场景。目标场景中可能包括各种类别的目标，例如物体、标志、建筑、行人、车辆等。本公开对目标场景的类型及目标的类别不作限制。
在一种可能的实现方式中，在对目标场景中的目标进行检测时，可通过传感设备采集目标场景的传感数据，以便对传感数据中的目标进行分析。在进行三维目标检测的情况下，传感设备可例如包括激光雷达、红绿蓝深度(Red Green Blue-Depth，RGB-D)采集设备等，采集到的传感数据可包括点云数据、RGB-D图像数据等。本公开对传感设备 的类型及采集到的传感数据的具体类型不作限制。
在一种可能的实现方式中，可连续采集到目标场景的多帧传感数据，通过电子设备依次对各帧传感数据进行目标检测。其中，如果传感数据为点云数据，则可直接进行处理；如果传感数据为RGB-D图像数据，则可将RGB-D图像数据进行反投影转换，得到点云数据后进行处理。
在一种可能的实现方式中，对于多帧点云数据中的第1帧，可在步骤S11中直接对第1帧点云数据进行目标检测，得到第1帧点云数据中目标的第一候选框；并在步骤S12中直接对第一候选框进行融合，得到第1帧点云数据的目标的第一检测框。
在一种可能的实现方式中，对于多帧点云数据中的第t帧(t为大于1的整数)，可在步骤S11中对第t帧点云数据进行目标检测，确定第t帧点云数据中目标的第一候选框。第一候选框的信息可包括第一候选框的中心点三维坐标(x 0,y 0,z 0)、长度、宽度、高度及旋转角度等信息。
在一种可能的实现方式中，可通过预训练的目标检测网络实现目标检测的过程，该目标检测网络可例如包括卷积神经网络(Convolutional Neural Networks，CNN)及区域生成网络(Region Proposal Network，RPN)，本公开对目标检测网络的具体网络结构不作限制。
在一种可能的实现方式中，在步骤S11和S12之前，在得到第t-1帧点云数据的检测结果后，可根据第t帧点云数据之前的t-1帧点云数据的检测结果，对前t-1帧点云数据中已经检测到的目标，在第t帧点云数据中的位置进行预测，得到这些目标在第t帧点云数据中的预测候选框。
在一种可能的实现方式中，在步骤S12中，根据第t帧点云数据的第一候选框和预测候选框，可确定出各第一候选框所对应的目标。例如根据各第一候选框与各预测候选框的交并比，对第一候选框和预测候选框进行匹配；对于存在匹配的预测候选框的第一候选框，将相匹配的预测候选框所对应的目标，确定为该第一候选框所对应的目标；对于不存在匹配的预测候选框的第一候选框，确定该第一候选框对应新的目标。
在一种可能的实现方式中，对于任意目标，可根据该目标的第一候选框及第一候选框对应的区域点云数据，与该目标的预测候选框及预测候选框对应的区域点云数据，进行候选框融合处理，从而确定出目标的实际检测框(可称为第一检测框)。
在一种可能的实现方式中，可通过预训练的融合网络实现候选框融合，也即将目标的第一候选框及第一候选框对应的区域点云数据，与该目标的预测候选框及预测候选框对应的区域点云数据，输入融合网络中处理，输出目标的第一检测框。该融合网络可例如包括区域卷积神经网络(Region Convolutional Neural Networks，RCNN)，本公开对融合网络的具体网络结构不作限制。
在一种可能的实现方式中，在对第t帧点云数据中的所有目标进行处理后，可得到第t帧点云数据的第一检测结果，该第一检测结果包括第t帧点云数据中各目标的第一检测框。
根据本公开的实施例，能够检测出第t帧点云数据中目标的第一候选框；通过历史检测结果预测得到的预测候选框对第一候选框进行修正，得到第t帧点云数据的检测结果，从而提高目标检测的精度。
在一种可能的实现方式中，步骤S11可包括：
根据所述第t帧点云数据中目标的预测概率图，将所述第t帧点云数据划分为存在目标的第一区域、不存在目标的第二区域以及未确定是否存在目标的第三区域；
对所述第t帧点云数据的第一区域及第三区域进行目标检测，确定所述第t帧点云数据中目标的第一候选框。
举例来说，在得到第t-1帧点云数据的检测结果后，可根据前t-1帧点云数据的检测结 果，预测得到针对第t帧点云数据中目标的预测候选框。根据该预测候选框，可预测第t帧点云数据的各个位置出现目标的概率，得到第t帧点云数据中目标的预测概率图。
在一种可能的实现方式中，可预先设置有第一概率阈值和第二概率阈值，第二概率阈值小于第一概率阈值。对于点云数据中的任意位置，如果该位置出现目标的概率大于第一概率阈值，则可认为该位置存在目标；如果该位置出现目标的概率小于第二概率阈值，则可认为该位置不存在目标；如果该位置出现目标的概率处于第一概率阈值与第二概率阈值之间，则不确定该位置是否存在目标，例如未检测过的位置，或检测过但仍未确定是否存在目标的位置。本公开对第一概率阈值和第二概率阈值的具体取值不作限制。
在一种可能的实现方式中，根据第t帧点云数据中目标的预测概率图，可基于第一概率阈值和第二概率阈值，将第t帧点云数据划分为存在目标的第一区域、不存在目标的第二区域以及未确定是否存在目标的第三区域。
经划分后，第二区域中不存在目标，可不对第二区域的点云数据进行目标检测。也即，对所述第t帧点云数据的第一区域及第三区域进行目标检测，确定第t帧点云数据中目标的第一候选框。
通过这种方式，可以减少目标检测所处理的点云数据的数据量，提高检测速度。
在一种可能的实现方式中，对所述第t帧点云数据的第一区域及第三区域进行目标检测，确定第t帧点云数据中目标的第一候选框的步骤，可包括：
对所述第一区域及所述第三区域的点云数据进行特征提取，得到第一点云特征；
对所述第一点云特征进行目标检测，确定所述第t帧点云数据中目标的第二候选框；
根据各个第二候选框的置信度，从所述第二候选框中确定出预设数量的第一候选框。
举例来说，可将第一区域及第三区域的点云数据输入目标检测网络的特征提取网络进行特征提取，得到点云数据的第一点云特征。该特征提取网络例如包括多个卷积层，本公开对特征提取网络的结构不作限制。
在一种可能的实现方式中，在特征提取之前，还可对第一区域及第三区域的点云数据进行采样，以减少处理的数据量。例如，通过随机采样，将具有N个点的点云数据采样为具有N/4个点的点云数据。将采样后的点云数据输入特征提取网络中处理，得到第一点云特征。通过这种方式，可以进一步提高检测速度。
在一种可能的实现方式中，可将第一点云特征输入目标检测网络的区域生成网络RPN进行处理，得到第t帧点云数据中目标的第二候选框。
在一种可能的实现方式中，第二候选框的数量较大，可进一步进行处理。根据各个第二候选框的置信度，可例如通过非最大抑制(Non-maximum suppression，NMS)方式，从第二候选框中确定出预设数量的第一候选框。该预设数量可例如取值为50个，本公开对此不作限制。
通过这种方式，可以点云数据中初步估计出目标对应的候选框，以便进行后续的处理。
在一种可能的实现方式中，步骤S11可包括：
对所述第t帧点云数据进行特征提取，得到第二点云特征；
对所述第二点云特征进行目标检测，确定所述第t帧点云数据中目标的第四候选框；
根据各个第四候选框的置信度，从所述第四候选框确定出预设数量的第一候选框。
举例来说，在未对第t帧点云数据划分区域的情况下，可直接对第t帧点云数据进行目标检测。将第t帧点云数据输入目标检测网络的特征提取网络进行特征提取，可得到第t帧点云数据的第二点云特征。该特征提取网络例如包括多个卷积层，本公开对特征提取网络的结构不作限制。
在一种可能的实现方式中，在特征提取之前，还可对第t帧点云数据进行采样，以减少处理的数据量。例如，通过随机采样，将具有M个点的点云数据采样为具有M/4个点的 点云数据。将采样后的点云数据输入特征提取网络中处理，得到第二点云特征。通过这种方式，可以进一步提高检测速度。
在一种可能的实现方式中，可将第二点云特征输入目标检测网络的区域生成网络RPN进行处理，得到第t帧点云数据中目标的第四候选框。
在一种可能的实现方式中，第四候选框的数量较大，可进一步进行处理。根据各个第四候选框的置信度，可例如通过非最大抑制(Non-maximum suppression，NMS)方式，从第四候选框中确定出预设数量的第一候选框。该预设数量可例如取值为50个，本公开对此不作限制。
通过这种方式，可以点云数据中初步估计出目标对应的候选框，以便进行后续的处理。
在一种可能的实现方式中，步骤S12可包括：
对所述第t帧点云数据中各个目标的预测候选框分别进行扩展，确定各个目标的第三候选框；
对所述第三候选框与所述第一候选框分别进行匹配，确定与各个第一候选框对应的目标；
根据所述第一候选框及与所述第一候选框所在区域对应的第一区域点云数据，以及所述第三候选框及与所述第三候选框所在区域对应的第二区域点云数据，对所述第t帧点云数据中的各个目标分别进行候选框融合，得到所述第t帧点云数据中各个目标的第一检测框。
举例来说，在对第t帧点云数据进行预测时，为第t帧点云数据的第一区域中的目标均预测出一个预测候选框，也即第一区域中的每个目标对应一个预测候选框。在步骤S12的处理中，可先对各个目标的预测候选框分别进行扩展，以便增加候选框的数量。
在一种可能的实现方式中，根据目标在第t帧点云数据中的预测候选框，可确定目标的位姿和尺度；根据目标的位姿和尺度的概率分布，可以以一定的方差和均值进行采样，扩展得到该目标的多个第三候选框。这样，可减小预测候选框的误差对后续处理的影响，提高与第一候选框匹配的概率，从而提高检测结果的稳定性，提高检测精度。
在一种可能的实现方式中，可对第三候选框与第一候选框分别进行匹配，确定与各个第一候选框对应的目标。其中，该步骤可包括：
分别确定各个第三候选框与各个第一候选框之间的交并比；
将与第一候选框的交并比大于或等于交并比阈值的第三候选框，确定为与第一候选框相匹配的第三候选框；
将与第一候选框相匹配的第三候选框对应的目标，确定为与所述第一候选框对应的目标。
也就是说，可通过交并比对第三候选框与第一候选框进行匹配。可分别确定各个第三候选框与各个第一候选框之间的交并比(Intersection-over-Union，IoU)。可预设有交并比阈值(例如0.5)，对于任意一个第一候选框，如果存在与该第一候选框之间的交并比大于或等于交并比阈值的第三候选框，则将该第三候选框确定为与该第一候选框相匹配的候选框；并将与该第三候选框对应的目标，确定为与该第一候选框对应的目标。将与该第三候选框对应的目标的标识(Identity Document，ID)，赋予该第一候选框，也即，认为相匹配的两个候选框对应于同一目标。
在一种可能的实现方式中，对于任意一个第一候选框，如果不存在与该第一候选框之间的交并比大于或等于交并比阈值的第三候选框，则可认为与第一候选框对应的目标为之前未出现过的新目标。在该情况下，可为与第一候选框对应的目标赋予新的ID。
通过这种方式，可以确定各个第一候选框的对应目标的标识，以便对相同标识的目标的候选框进行融合。
在一种可能的实现方式中，根据所述第一候选框及与所述第一候选框所在区域对应的第一区域点云数据，以及所述第三候选框及与所述第三候选框所在区域对应的第二区域点云数据，对所述第t帧点云数据中的各个目标分别进行候选框融合，得到所述第t帧点云数据中各个目标的第一检测框。
在一种可能的实现方式中，对于第t帧点云数据中的任意一个目标，如果该目标存在第一候选框和第三候选框，则可从第t帧点云数据中分割出与该目标的第一候选框所在区域对应的第一区域点云数据，并分割出与该目标的第三候选框所在区域对应的第二区域点云数据。将该目标的第一候选框及第一区域点云数据，第三候选框及第二区域点云数据，输入到预训练的融合网络中处理，输出该目标的第一检测框。该第一检测框包括三维的区域框。
在一种可能的实现方式中，对于第t帧点云数据中的任意一个目标，如果该目标仅存在第一候选框，则可从第t帧点云数据中分割出与该目标的第一候选框所在区域对应的第一区域点云数据。将该目标的第一候选框及第一区域点云数据，输入到预训练的融合网络中处理，输出该目标的第一检测框。
在一种可能的实现方式中，对第t帧点云数据中的所有目标进行上述处理，可得到第t帧点云数据中所有目标的第一检测框。
在一种可能的实现方式中，可将第t帧点云数据中所有目标的第一检测框作为第t帧点云数据的检测结果(可称为第一检测结果)；也可进行其他处理(例如对目标进行分类)，以使第t帧点云数据的检测结果包括更多的内容。本公开对此不作限制。
通过这种方式，可确定第t帧点云数据中所有目标的第一检测框，实现第t帧点云数据中目标的精确检测。
在一种可能的实现方式中，所述第一检测结果还包括所述第t帧点云数据中目标的类别，步骤S12包括：
根据与第二目标的第一检测框所在区域对应的第三区域点云数据，对所述第二目标进行分类，确定所述第二目标的类别，所述第二目标为所述第t帧点云数据中的任意一个目标。
举例来说，可以在步骤S12中对第t帧点云数据中的目标进行分类。对于第t帧点云数据中的任意一个目标(可称为第二目标)，可根据该第二目标的第一检测框，从第t帧点云数据中分割出于第一检测框所在区域对应的第三区域点云数据。
在一种可能的实现方式中，可将该第三区域点云数据输入预训练的分类网络中处理，确定出第二目标所属的类别。该分类网络可例如包括卷积层、全连接层等，本公开对分类网络的具体网络结构不作限制。
在一种可能的实现方式中，对第t帧点云数据中的所有目标进行上述处理，可得到第t帧点云数据中所有目标的类别，从而将目标的类别加入第t帧点云数据的第一检测结果。
通过这种方式，可以检测出的目标信息更为丰富。
在步骤S12中得到第t帧点云数据的第一检测结果后，还可以将该第一检测结果与之前的历史检测结果相结合，进一步优化第t帧点云数据的检测结果。
在一种可能的实现方式中，根据本公开实施例的目标检测方法还可包括：
首先，获取在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果；然后，根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果。
也就是说，前t-1帧点云数据已经在之前的处理中得到了最终检测结果(可称为第二检测结果)，每个第二检测结果包括目标的第二检测框，第t帧点云数据中的目标可能在t-1帧点云数据的第二检测结果中存在对应的第二检测框。
在一种可能的实现方式中，对于第t帧点云数据中的任意一个目标，如果在前t-1帧点 云数据的第二检测结果中存在该目标的第二检测框，则可根据该目标在前t-1帧点云数据中的第二检测框，对该目标在第t帧点云数据中的第一检测框进行修正，得到修正后的检测框，称为第二检测框。
在一种可能的实现方式中，如果在前t-1帧点云数据的第二检测结果中不存在该目标的第二检测框，则可将该目标在第t帧点云数据中的第一检测框直接作为第二检测框。
在一种可能的实现方式中，对第t帧点云数据中的所有目标进行上述处理，可得到第t帧点云数据中所有目标的第二检测框，从而得到第t帧点云数据的第二检测结果。
通过这种方式，可以进一步提升目标检测的精度。
在一种可能的实现方式中，所述根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果的步骤，可包括：
确定第一目标的检测框集合，所述第一目标为所述第t帧点云数据中的任意一个目标，所述第一目标的检测框集合包括所述第一目标在t-1帧点云数据的第二检测结果中的第二检测框，以及所述第一目标在第t帧点云数据的第一检测结果中的第一检测框；
对于所述第一目标的检测框集合中任意一个检测框，将所述检测框集合中与所述检测框之间的误差小于或等于误差阈值的检测框，确定为所述检测框的内点框；
从所述第一目标的检测框集合中确定出内点框数量最多的第三检测框；
对所述第三检测框及所述第三检测框的所有内点框进行融合，确定所述第t帧点云数据中第一目标的第二检测框。
举例来说，对于第t帧点云数据中的任意一个目标(称为第一目标)，可获取该第一目标的检测框集合。该检测框集合中包括第一目标在t-1帧点云数据的第二检测结果中的第二检测框，以及第一目标在第t帧点云数据的第一检测结果中的第一检测框。
在一种可能的实现方式中，对于第一目标的检测框集合中任意一个检测框，可确定检测框集合中的其他检测框与该检测框之间的误差。可预设有误差阈值，与该检测框之间的误差小于或等于该误差阈值的检测框，可确定为该检测框的内点框；反之，与该检测框之间的误差大于该误差阈值的检测框，可确定为该检测框的外点框。本公开对误差阈值的具体取值不作限制。
在一种可能的实现方式中，可从第一目标的检测框集合中确定出内点框数量最多的第三检测框，将该第三检测框作为初始估计的检测框。对第三检测框及第三检测框的所有内点框进行融合优化，可得到第一目标的位置信息的最优估计，也即得到修正后的第二检测框。
在一种可能的实现方式中，可通过最小二乘的方式对第三检测框及第三检测框的所有内点框进行对融合优化，也可通过卡尔曼滤波的方式对第三检测框及第三检测框的所有内点框进行对融合优化，本公开对融合优化的具体方式不作限制。
在一种可能的实现方式中，对第t帧点云数据中的所有目标进行上述处理，可得到第t帧点云数据中所有目标的第二检测框，从而得到第t帧点云数据的第二检测结果。
通过这种方式，可以将检测结果与之前的历史检测结果相结合，进一步优化第t帧点云数据的检测结果，提高目标检测的精度。
在一种可能的实现方式中，所述方法还包括：
根据所述t-1帧点云数据的第二检测结果，以及所述第t帧点云数据的第二检测结果，对第t+1帧点云数据中目标的运动状态进行预测，确定所述第t+1帧点云数据中目标的预测候选框。
举例来说，在得到第t帧点云数据的第二检测结果后，可根据历史检测结果对第t+1帧点云数据进行预测，帮助第t+1帧点云数据的目标检测。
在一种可能的实现方式中，对于第t帧点云数据中的任意一个目标(可称为第三目 标)，可获取该第三目标在t帧点云数据的第二检测结果中的第二检测框。如果该第三目标存在多个第二检测框，则可根据相邻帧的各个第二检测框之间的误差，对第t+1帧点云数据中目标的运动状态进行预测，预测出该第三目标在第t+1帧点云数据中的位置，得到该第三目标在第t+1帧点云数据中的预测候选框。
在一种可能的实现方式中，可通过卡尔曼滤波或最小二乘的方式实现运动状态的预测，本公开对此不作限制。
在一种可能的实现方式中，如果该第三目标仅存在一个第二检测框，也即第三目标为第t帧点云数据中新出现的目标，则可根据该第三目标附近的其他目标进行预测，通过其他目标在第t帧点云数据中的第二检测框及在第t+1帧点云数据中的预测候选框之间的误差，来预测得到该第三目标在第t+1帧点云数据中的预测候选框。
这样，对第t帧点云数据中的所有目标进行预测，可确定出第t+1帧点云数据的已检测区域中所有目标的预测候选框。
通过这种方式，可以得到第t+1帧点云数据中目标的预测候选框，帮助第t+1帧点云数据的目标检测，从而提高检测精度。
在一种可能的实现方式中，所述方法还包括：
根据所述第t+1帧点云数据中目标的预测候选框以及第t帧点云数据，对所述第t帧点云数据中目标的预测概率图进行更新，确定所述第t+1帧点云数据中目标的预测概率图。
举例来说，在得到第t+1帧点云数据中目标的预测候选框后，可根据该预测候选框以及第t帧点云数据，对第t帧点云数据中目标的预测概率图进行更新。也即，根据目标在第t帧点云数据中的位置和第t+1帧点云数据中的位置(预测候选框)，确定预测概率图中的各个位置是否存在目标，更新各个位置可能出现目标的概率，从而得到第t+1帧点云数据中目标的预测概率图。
通过这种方式，可得到第t+1帧点云数据中目标的预测概率图，以便在后续处理时为第t+1帧点云数据划分多个区域，从而提高目标检测的速度。
在一种可能的实现方式中，所述方法还包括：
根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据中目标的运动状态进行预测，确定所述第t帧点云数据中目标的预测候选框。
也就是说，在得到第t-1帧点云数据的第二检测结果后，可根据历史检测结果对第t帧点云数据进行预测，帮助第t帧点云数据的目标检测。对于第t-1帧点云数据中的任意一个目标，可获取该目标在前t-1帧点云数据的第二检测结果中的第二检测框，对第t帧点云数据中目标的运动状态进行预测，预测出该目标在第t帧点云数据中的位置，得到该目标在第t帧点云数据中的预测候选框。该预测过程与对第t+1帧点云数据的预测过程类似，此处不再重复描述。
通过这种方式，可以得到第t帧点云数据中目标的预测候选框，帮助第t帧点云数据的目标检测，从而提高检测精度。
在一种可能的实现方式中，所述方法还包括：
根据第t帧点云数据中目标的预测候选框，以及第t-1帧点云数据，对所述第t-1帧点云数据中目标的预测概率图进行更新，确定所述第t帧点云数据中目标的预测概率图。
也就是说，在得到第t帧点云数据中目标的预测候选框后，可根据该预测候选框以及第t-1帧点云数据，对第t-1帧点云数据中目标的预测概率图进行更新，得到第t帧点云数据中目标的预测概率图。该更新过程与对第t+1帧点云数据的预测概率图的更新过程类似，此处不再重复描述。
通过这种方式，可得到第t帧点云数据中目标的预测概率图，以便在后续处理时为第t帧点云数据划分多个区域，从而提高目标检测的速度。
在本公开实施例中，可以通过如图1B所示的网络架构，实现对待重建对象的三维重 建，图1B示出本公开实施例目标检测方法的一种网络架构示意图，该网络架构中包括：用户终端201、网络202和目标检测终端203。为实现支撑一个示例性应用用户终端201和目标检测203通过网络202建立有通信连接，用户终端201需要对包括待检测目标的目标场景的第t帧点云数据进行目标检测时，首先，将目标场景的第t帧点云数据通过网络202发送至目标检测终端203；然后，目标检测终端203通过对该目标的检测框进行预测，得到第一候选框；最后，目标检测终端203通过该第一候选框，实现对第t帧点云数据中目标进行检测，得到该目标的检测结果。如此，通过多帧点云数据进行检测框的预测，能够提高目标检测的精度。
图2示出根据本公开实施例的目标检测方法的处理过程的示意图。如图2所示，可将对当前帧进行目标检测处理的过程称为前端；将记录有历史结果，并根据历史结果对当前帧进行修正、对下一帧进行预测的过程称为后端，后端的处理也可称为目标追踪与融合。其中，当前帧为第t帧。
在示例中，之前的第t-1帧的前端处理中得到了第t-1帧点云数据的第一检测结果(未示出)；将该第一检测结果与前t-2帧的历史检测结果关联，在第t-1帧后端的步骤211中通过卡尔曼滤波或最小二乘的方式进行检测框的融合优化，实现检测结果的修正，得到第t-1帧点云数据的第二检测结果(未示出)。
在示例中，在第t-1帧的后端处理中，可根据前t-1帧的历史检测结果，对第t帧中的目标进行运动预测212，得到第t帧点云数据中目标的预测候选框213；再根据预测候选框213及第t-1帧点云数据(未示出)，在步骤214中对第t-1帧的预测概率图进行更新，得到第t帧点云数据中目标的预测概率图215，从而完成了第t-1帧的整个处理过程。
在示例中，在第t帧的前端处理中，可根据预测概率图215，将所第t帧点云数据221划分为存在目标的第一区域、不存在目标的第二区域以及未确定是否存在目标的第三区域，得到划分区域后的点云数据222。将点云数据222的第一区域及第三区域输入到目标检测网络223中进行目标检测，可得到预设数量的第一候选框。将第t帧点云数据中目标的预测候选框213与第一候选框进行匹配，确定各个第一候选框所对应的目标标识，得到待处理的所有候选框224(每个目标对应多个框)。将目标的所有候选框224及候选框224对应的区域点云数据，输入到融合网络225中处理，得到目标的第一检测框(每个目标对应一个框)，作为第t帧点云数据的第一检测结果226。并可在步骤227中将第一检测结果226与前t-1帧的历史检测结果关联。
在示例中，在第t帧的后端处理中，可在步骤231中，通过卡尔曼滤波或最小二乘的方式进行检测框的融合优化，实现检测结果的修正，得到第t帧点云数据中各目标的第二检测框，作为第t帧点云数据的第二检测结果230，也即最终的输出结果。
在示例中，在第t帧的后端处理中，可根据前t帧的第二检测结果，对第t+1帧中的目标进行运动预测232，得到第t+1帧点云数据中目标的预测候选框233；再根据预测候选框233及第t帧点云数据221，在步骤234中对第t帧的预测概率图215进行更新，得到第t+1帧点云数据中目标的预测概率图235，从而完成了第t帧的整个处理过程。
图3a示出目标场景的图像的示意图；图3b示出目标的检测结果的示意图。如图3a所示，目标场景中包括多个椅子，椅子可作为待检测的目标。如图3b所示，检测框31为根据相关技术的单帧处理的目标检测方法得到的检测结果；检测框32为目标的真实三维图像框；检测框33为根据本公开实施例的目标检测方法得到的检测结果。
可见，本公开实施例的目标检测方法得到的检测结果的精度较高。在目标被部分遮挡的情况下，相关技术的检测结果明显变差，而本公开实施例的目标检测方法仍然能够保持较高的精度。
根据本公开的实施例的目标检测方法，在对目标场景的连续多帧点云数据进行三维目标检测的情况下，能够有效利用历史检测结果进行三维目标的检测与追踪；能够通过 历史检测结果预测当前帧中目标的候选框，和当前帧中已知区域可能出现3D物体的概率的分布图，并反馈到当前帧的目标检测过程中；能够使得当前帧在目标检测时，利用预测的概率分布图划分区域，从而减少处理的数据量，提高目标检测的速度；并利用预测的候选框作为先验框，不但避免每一帧对整个场景进行目标搜索，还根据先验框得到更准确的候选框，有效提高了目标检测的精度，并避免漏检情况的发生。
根据本公开的实施例的目标检测方法，能够进行目标的追踪与融合，将每个3D目标在连续时间上所有的检测框都储存为该3D物体的历史检测框，在每一帧都分别对每一个3D目标的所有历史检测框进行融合与优化，以获取当前帧的3D目标的位置的最优估计，从而有效提升3D检测框的稳定性，减小目标被遮挡或截断时的检测误差，显著提高目标检测的精度和鲁棒性。
根据本公开的实施例的目标检测方法，能够应用于增强现实AR、室内导航等应用场景中，实现3D目标的估计与检测。相关技术的处理方式没有考虑同一物体在连续帧内位置信息的关系，没有利用到连续时间上的信息，容易造成3D检测框的抖动。例如在室内场景下，由于物体尺度更大，检测框抖动的现象也会更加严重。而根据本公开的实施例的目标检测方法，通过利用连续帧内位置信息的关系及连续时间上的信息，能够输出更为稳定的3D检测框，减小检测误差。
可以理解，本公开提及的上述各个方法实施例，在不违背原理逻辑的情况下，均可以彼此相互结合形成结合后的实施例，限于篇幅，本公开不再赘述。本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
此外，本公开还提供了目标检测装置、电子设备、计算机可读存储介质、程序，上述均可用来实现本公开提供的任一种目标检测方法，相应技术方案和描述和参见方法部分的相应记载，不再赘述。
图4示出根据本公开实施例的目标检测装置的框图，如图4所示，所述装置包括：
第一检测模块41，配置为对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，t为大于1的整数；
第二检测模块42，配置为根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框，其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的。
在一种可能的实现方式中，所述第一检测模块包括：区域划分子模块，配置为根据所述第t帧点云数据中目标的预测概率图，将所述第t帧点云数据划分为存在目标的第一区域、不存在目标的第二区域以及未确定是否存在目标的第三区域；第一检测子模块，配置为对所述第一区域及所述第三区域进行目标检测，确定所述第t帧点云数据中目标的第一候选框。
在一种可能的实现方式中，所述装置还包括：修正模块，配置为获取在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果；并根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果。
在一种可能的实现方式中，所述装置还包括：第一运动预测模块，配置为根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据中目标的运动状态进行预测，确定所述第t帧点云数据中目标的预测候选框。
在一种可能的实现方式中，所述装置还包括：第一概率图更新模块，配置为根据所述第t帧点云数据中目标的预测候选框，以及第t-1帧点云数据，对所述第t-1帧点云数据中目标的预测概率图进行更新，确定所述第t帧点云数据中目标的预测概率图。
在一种可能的实现方式中，所述第一检测子模块，配置为：对所述第一区域及所述第三区域的点云数据进行特征提取，得到第一点云特征；对所述第一点云特征进行目标检测，确定所述第t帧点云数据中目标的第二候选框；根据各个第二候选框的置信度，从所述第二候选框中确定出预设数量的第一候选框。
在一种可能的实现方式中，所述第二检测模块包括：候选框扩展子模块，配置为对所述第t帧点云数据中各个目标的预测候选框分别进行扩展，确定各个目标的第三候选框；候选框匹配子模块，配置为对所述第三候选框与所述第一候选框分别进行匹配，确定与各个第一候选框对应的目标；候选框融合子模块，配置为根据所述第一候选框及与所述第一候选框所在区域对应的第一区域点云数据，以及所述第三候选框及与所述第三候选框所在区域对应的第二区域点云数据，对所述第t帧点云数据中的各个目标分别进行候选框融合，得到所述第t帧点云数据中各个目标的第一检测框。
在一种可能的实现方式中，所述候选框匹配子模块，配置为：分别确定各个第三候选框与各个第一候选框之间的交并比；将与第一候选框的交并比大于或等于交并比阈值的第三候选框，确定为与第一候选框相匹配的第三候选框；将与第一候选框相匹配的第三候选框对应的目标，确定为与所述第一候选框对应的目标。
在一种可能的实现方式中，每个第二检测结果包括目标的第二检测框，所述修正模块包括：集合确定子模块，配置为确定第一目标的检测框集合，所述第一目标为所述第t帧点云数据中的任意一个目标，所述第一目标的检测框集合包括所述第一目标在t-1帧点云数据的第二检测结果中的第二检测框，以及所述第一目标在第t帧点云数据的第一检测结果中的第一检测框；内点框确定子模块，配置为对于所述第一目标的检测框集合中任意一个检测框，将所述检测框集合中与所述检测框之间的误差小于或等于误差阈值的检测框，确定为所述检测框的内点框；检测框选择子模块，配置为从所述第一目标的检测框集合中确定出内点框数量最多的第三检测框；内点框融合子模块，配置为对所述第三检测框及所述第三检测框的所有内点框进行融合，确定所述第t帧点云数据中第一目标的第二检测框。
在一种可能的实现方式中，所述装置还包括：第二运动预测模块，配置为根据所述t-1帧点云数据的第二检测结果，以及所述第t帧点云数据的第二检测结果，对第t+1帧点云数据中目标的运动状态进行预测，确定所述第t+1帧点云数据中目标的预测候选框。
在一种可能的实现方式中，所述装置还包括：第二概率图更新模块，配置为根据所述第t+1帧点云数据中目标的预测候选框，以及第t帧点云数据，对所述第t帧点云数据中目标的预测概率图进行更新，确定所述第t+1帧点云数据中目标的预测概率图。
在一种可能的实现方式中，所述第一检测模块包括：特征提取子模块，配置为对所述第t帧点云数据进行特征提取，得到第二点云特征；第二检测子模块，配置为对所述第二点云特征进行目标检测，确定所述第t帧点云数据中目标的第四候选框；选择子模块，配置为根据各个第四候选框的置信度，从所述第四候选框确定出预设数量的第一候选框。
在一种可能的实现方式中，所述第一检测结果还包括所述第t帧点云数据中目标的类别，所述第二检测模块包括：分类子模块，配置为根据与第二目标的第一检测框所在区域对应的第三区域点云数据，对所述第二目标进行分类，确定所述第二目标的类别，所述第二目标为所述第t帧点云数据中的任意一个目标。
在一种可能的实现方式中，所述目标场景包括室内场景，所述第t帧点云数据中的目标包括物体，所述第t帧点云数据中目标的第一检测框包括三维区域框。
在一些实施例中，本公开实施例提供的装置具有的功能或包含的模块可以用于执行上文方法实施例描述的方法，其具体实现可以参照上文方法实施例的描述，为了简洁，这里不再赘述。
本公开实施例还提出一种计算机可读存储介质，其上存储有计算机程序指令，所述 计算机程序指令被处理器执行时实现上述方法。计算机可读存储介质可以是非易失性计算机可读存储介质。
本公开实施例还提出一种电子设备，包括：处理器；用于存储处理器可执行指令的存储器；其中，所述处理器被配置为调用所述存储器存储的指令，以执行上述方法。
本公开实施例还提供了一种计算机程序产品，包括计算机可读代码，当计算机可读代码在设备上运行时，设备中的处理器执行用于实现如上任一实施例提供的目标检测方法的指令。
本公开实施例还提供了另一种计算机程序产品，用于存储计算机可读指令，指令被执行时使得计算机执行上述任一实施例提供的目标检测方法的操作。
电子设备可以被提供为终端、服务器或其它形态的设备。
图5示出根据本公开实施例的一种电子设备800的框图。例如，电子设备800可以是移动电话，计算机，数字广播终端，消息收发设备，游戏控制台，平板设备，医疗设备，健身设备，个人数字助理等终端。
参照图5，电子设备800可以包括以下一个或多个组件：处理组件802，存储器804，电源组件806，多媒体组件808，音频组件810，输入/输出(I/O)的接口812，传感器组件814，以及通信组件816。
处理组件802通常控制电子设备800的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联的操作。处理组件802可以包括一个或多个处理器820来执行指令，以完成上述的方法的全部或部分步骤。此外，处理组件802可以包括一个或多个模块，便于处理组件802和其他组件之间的交互。例如，处理组件802可以包括多媒体模块，以方便多媒体组件808和处理组件802之间的交互。
存储器804被配置为存储各种类型的数据以支持在电子设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令，联系人数据，电话簿数据，消息，图片，视频等。存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(SRAM)，电可擦除可编程只读存储器(EEPROM)，可擦除可编程只读存储器(EPROM)，可编程只读存储器(PROM)，只读存储器(ROM)，磁存储器，快闪存储器，磁盘或光盘。
电源组件806为电子设备800的各种组件提供电力。电源组件806可以包括电源管理系统，一个或多个电源，及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中，屏幕可以包括液晶显示器(LCD)和触摸面板(TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中，多媒体组件808包括一个前置摄像头和/或后置摄像头。当电子设备800处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如，音频组件810包括一个麦克风(MIC)，当电子设备800处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在存储器804或经由通信组件816发送。在一些实施例中，音频组件810还包括一个扬声器，用于输出音频信号。
I/O接口812为处理组件802和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和 锁定按钮。
传感器组件814包括一个或多个传感器，用于为电子设备800提供各个方面的状态评估。例如，传感器组件814可以检测到电子设备800的打开/关闭状态，组件的相对定位，例如所述组件为电子设备800的显示器和小键盘，传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变，用户与电子设备800接触的存在或不存在，电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器，如互补金属氧化物半导体(CMOS)或电荷耦合装置(CCD)图像传感器，用于在成像应用中使用。在一些实施例中，该传感器组件814还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络，如无线网络(WiFi)，第二代移动通信技术(2G)或第三代移动通信技术(3G)，或它们的组合。在一个示例性实施例中，通信组件816经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在一个示例性实施例中，所述通信组件816还包括近场通信(NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(RFID)技术，红外数据协会(IrDA)技术，超宽带(UWB)技术，蓝牙(BT)技术和其他技术来实现。
在示例性实施例中，电子设备800可以被一个或多个应用专用集成电路(ASIC)、数字信号处理器(DSP)、数字信号处理设备(DSPD)、可编程逻辑器件(PLD)、现场可编程门阵列(FPGA)、控制器、微控制器、微处理器或其他电子元件实现，用于执行上述方法。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器804，上述计算机程序指令可由电子设备800的处理器820执行以完成上述方法。
图6示出根据本公开实施例的一种电子设备1900的框图。例如，电子设备1900可以被提供为一服务器。参照图6，电子设备1900包括处理组件1922，其进一步包括一个或多个处理器，以及由存储器1932所代表的存储器资源，用于存储可由处理组件1922的执行的指令，例如应用程序。存储器1932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件1922被配置为执行指令，以执行上述方法。
电子设备1900还可以包括一个电源组件1926被配置为执行电子设备1900的电源管理，一个有线或无线网络接口1950被配置为将电子设备1900连接到网络，和一个输入输出(I/O)接口1958。电子设备1900可以操作基于存储在存储器1932的操作系统，例如微软服务器操作系统(Windows ServerTM)，苹果公司推出的基于图形用户界面操作系统(Mac OS XTM)，多用户多进程的计算机操作系统(UnixTM),自由和开放原代码的类Unix操作系统(LinuxTM)，开放原代码的类Unix操作系统(FreeBSDTM)或类似。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器1932，上述计算机程序指令可由电子设备1900的处理组件1922执行以完成上述方法。
本公开可以是系统、方法和/或计算机程序产品。计算机程序产品可以包括计算机可读存储介质，其上载有用于使处理器实现本公开的各个方面的计算机可读程序指令。
计算机可读存储介质可以是可以保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是但不限于电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、静态随机存取存储器 (SRAM)、便携式压缩盘只读存储器(CD-ROM)、数字多功能盘(DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身，诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(例如，通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里所描述的计算机可读程序指令可以从计算机可读存储介质下载到各个计算/处理设备，或者通过网络、例如因特网、局域网、广域网和/或无线网下载到外部计算机或外部存储设备。网络可以包括铜传输电缆、光纤传输、无线传输、路由器、防火墙、交换机、网关计算机和/或边缘服务器。每个计算/处理设备中的网络适配卡或者网络接口从网络接收计算机可读程序指令，并转发该计算机可读程序指令，以供存储在各个计算/处理设备中的计算机可读存储介质中。
用于执行本公开操作的计算机程序指令可以是汇编指令、指令集架构(ISA)指令、机器指令、机器相关指令、微代码、固件指令、状态设置数据、或者以一种或多种编程语言的任意组合编写的源代码或目标代码，所述编程语言包括面向对象的编程语言—诸如Smalltalk、C++等，以及常规的过程式编程语言—诸如“C”语言或类似的编程语言。计算机可读程序指令可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络—包括局域网(LAN)或广域网(WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。在一些实施例中，通过利用计算机可读程序指令的状态信息来个性化定制电子电路，例如可编程逻辑电路、现场可编程门阵列(FPGA)或可编程逻辑阵列(PLA)，该电子电路可以执行计算机可读程序指令，从而实现本公开的各个方面。
这里参照根据本公开实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本公开的各个方面。应当理解，流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合，都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器，从而生产出一种机器，使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时，产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中，这些指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作，从而，存储有指令的计算机可读介质则包括一个制造品，其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
也可以把计算机可读程序指令加载到计算机、其它可编程数据处理装置、或其它设备上，使得在计算机、其它可编程数据处理装置或其它设备上执行一系列操作步骤，以产生计算机实现的过程，从而使得在计算机、其它可编程数据处理装置、或其它设备上执行的指令实现流程图和/或框图中的一个或多个方框中规定的功能/动作。
附图中的流程图和框图显示了根据本公开的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分，所述模块、程序段或指令的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或动作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指 令的组合来实现。
该计算机程序产品可以具体通过硬件、软件或其结合的方式实现。在一个可选实施例中，所述计算机程序产品具体体现为计算机存储介质，在另一个可选实施例中，计算机程序产品具体体现为软件产品，例如软件开发包(Software Development Kit，SDK)等等。
以上已经描述了本公开的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中的技术的改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。
工业实用性
本公开涉及一种目标检测方法及装置、电子设备和存储介质，所述方法包括：对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，t为大于1的整数；根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框，其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的。

Claims (17)

  1. 一种目标检测方法，包括：
    对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，t为大于1的整数；
    根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框；
    其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的。
  2. 根据权利要求1所述的方法，所述对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，包括：
    根据所述第t帧点云数据中目标的预测概率图，将所述第t帧点云数据划分为存在目标的第一区域、不存在目标的第二区域以及未确定是否存在目标的第三区域；
    对所述第一区域及所述第三区域进行目标检测，确定所述第t帧点云数据中目标的第一候选框。
  3. 根据权利要求1或2所述的方法，所述方法还包括：
    获取在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果；
    根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果。
  4. 根据权利要求1至3中任意一项所述的方法，所述方法还包括：
    根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据中目标的运动状态进行预测，确定所述第t帧点云数据中目标的预测候选框。
  5. 根据权利要求2所述的方法，所述方法还包括：
    根据所述第t帧点云数据中目标的预测候选框，以及第t-1帧点云数据，对所述第t-1帧点云数据中目标的预测概率图进行更新，确定所述第t帧点云数据中目标的预测概率图。
  6. 根据权利要求2或5所述的方法，所述对所述第一区域及所述第三区域进行目标检测，确定所述第t帧点云数据中目标的第一候选框，包括：
    对所述第一区域及所述第三区域的点云数据进行特征提取，得到第一点云特征；
    对所述第一点云特征进行目标检测，确定所述第t帧点云数据中目标的第二候选框；
    根据各个第二候选框的置信度，从所述第二候选框中确定出预设数量的第一候选框。
  7. 根据权利要求1至6中任意一项所述的方法，所述根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，包括：
    对所述第t帧点云数据中各个目标的预测候选框分别进行扩展，确定各个目标的第三候选框；
    对所述第三候选框与所述第一候选框分别进行匹配，确定与各个第一候选框对应的目标；
    根据所述第一候选框及与所述第一候选框所在区域对应的第一区域点云数据，以及所述第三候选框及与所述第三候选框所在区域对应的第二区域点云数据，对所述第t帧点云数据中的各个目标分别进行候选框融合，得到所述第t帧点云数据中各个目标的第一检测框。
  8. 根据权利要求7所述的方法，所述对所述第三候选框与所述第一候选框分别进行 匹配，确定与各个第一候选框对应的目标，包括：
    分别确定各个第三候选框与各个第一候选框之间的交并比；
    将与第一候选框的交并比大于或等于交并比阈值的第三候选框，确定为与第一候选框相匹配的第三候选框；
    将与第一候选框相匹配的第三候选框对应的目标，确定为与所述第一候选框对应的目标。
  9. 根据权利要求3所述的方法，每个第二检测结果包括目标的第二检测框，
    所述根据在所述第t帧点云数据之前的t-1帧点云数据的第二检测结果，对所述第t帧点云数据的第一检测结果进行修正，确定所述第t帧点云数据的第二检测结果，包括：
    确定第一目标的检测框集合，所述第一目标为所述第t帧点云数据中的任意一个目标，所述第一目标的检测框集合包括：所述第一目标在所述t-1帧点云数据的第二检测结果中的第二检测框，以及所述第一目标在第t帧点云数据的第一检测结果中的第一检测框；
    对于所述第一目标的检测框集合中任意一个检测框，将所述检测框集合中与所述检测框之间的误差小于或等于误差阈值的检测框，确定为所述检测框的内点框；
    从所述第一目标的检测框集合中确定出内点框数量最多的第三检测框；
    对所述第三检测框及所述第三检测框的所有内点框进行融合，确定所述第t帧点云数据中第一目标的第二检测框。
  10. 根据权利要求3或9所述的方法，所述方法还包括：
    根据所述t-1帧点云数据的第二检测结果，以及所述第t帧点云数据的第二检测结果，对第t+1帧点云数据中目标的运动状态进行预测，确定所述第t+1帧点云数据中目标的预测候选框。
  11. 根据权利要求10所述的方法，所述方法还包括：
    根据所述第t+1帧点云数据中目标的预测候选框，以及第t帧点云数据，对所述第t帧点云数据中目标的预测概率图进行更新，确定所述第t+1帧点云数据中目标的预测概率图。
  12. 根据权利要求1所述的方法，所述对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，包括：
    对所述第t帧点云数据进行特征提取，得到第二点云特征；
    对所述第二点云特征进行目标检测，确定所述第t帧点云数据中目标的第四候选框；
    根据各个第四候选框的置信度，从所述第四候选框确定出预设数量的第一候选框。
  13. 根据权利要求1至12中任意一项所述的方法，所述第一检测结果还包括所述第t帧点云数据中目标的类别，
    所述根据所述第t帧点云数据、所述第一候选框以及针对所述第t帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，包括：
    根据与第二目标的第一检测框所在区域对应的第三区域点云数据，对所述第二目标进行分类，确定所述第二目标的类别，所述第二目标为所述第t帧点云数据中的任意一个目标。
  14. 根据权利要求1至13中任意一项所述的方法，所述目标场景包括室内场景，所述第t帧点云数据中的目标包括物体，所述第t帧点云数据中目标的第一检测框包括三维区域框。
  15. 一种目标检测装置，包括：
    第一检测模块，配置为对目标场景的第t帧点云数据进行目标检测，确定所述第t帧点云数据中目标的第一候选框，t为大于1的整数；
    第二检测模块，配置为根据所述第t帧点云数据、所述第一候选框以及针对所述第t 帧点云数据中目标的预测候选框，确定所述第t帧点云数据的第一检测结果，所述第一检测结果包括所述第t帧点云数据中目标的第一检测框；
    其中，所述预测候选框是根据所述第t帧点云数据之前的t-1帧点云数据的检测结果预测得到的。
  16. 一种电子设备，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为调用所述存储器存储的指令，以执行权利要求1至14中任意一项所述的方法。
  17. 一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现权利要求1至14中任意一项所述的方法。
PCT/CN2021/078481 2020-07-28 2021-03-01 目标检测方法及装置、电子设备和存储介质 WO2022021872A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
KR1020227003199A KR20220027202A (ko) 2020-07-28 2021-03-01 객체 검출 방법 및 장치, 전자 기기 및 저장매체
JP2022505272A JP2022546201A (ja) 2020-07-28 2021-03-01 ターゲット検出方法および装置、電子機器並びに記憶媒体

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010738105.2A CN111881827B (zh) 2020-07-28 2020-07-28 目标检测方法及装置、电子设备和存储介质
CN202010738105.2 2020-07-28

Publications (1)

Publication Number Publication Date
WO2022021872A1 true WO2022021872A1 (zh) 2022-02-03

Family

ID=73200364

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/078481 WO2022021872A1 (zh) 2020-07-28 2021-03-01 目标检测方法及装置、电子设备和存储介质

Country Status (5)

Country Link
JP (1) JP2022546201A (zh)
KR (1) KR20220027202A (zh)
CN (1) CN111881827B (zh)
TW (1) TWI758205B (zh)
WO (1) WO2022021872A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111881827B (zh) * 2020-07-28 2022-04-26 浙江商汤科技开发有限公司 目标检测方法及装置、电子设备和存储介质
CN112529943B (zh) * 2020-12-22 2024-01-16 深圳市优必选科技股份有限公司 一种物体检测方法、物体检测装置及智能设备
CN113420725B (zh) * 2021-08-20 2021-12-31 天津所托瑞安汽车科技有限公司 Bsd产品的漏报场景识别方法、设备、系统和存储介质
CN113838125A (zh) * 2021-09-17 2021-12-24 中国第一汽车股份有限公司 目标位置确定方法、装置、电子设备以及存储介质
CN116052155A (zh) * 2021-10-27 2023-05-02 华为技术有限公司 一种点云数据处理方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109188457A (zh) * 2018-09-07 2019-01-11 百度在线网络技术（北京）有限公司 物体检测框的生成方法、装置、设备、存储介质及车辆
CN110728210A (zh) * 2019-09-25 2020-01-24 上海交通大学 一种三维点云数据的半监督目标标注方法和系统
WO2020128650A1 (en) * 2018-12-19 2020-06-25 Sony Corporation Point cloud coding structure
CN111427979A (zh) * 2020-01-15 2020-07-17 深圳市镭神智能系统有限公司 基于激光雷达的动态地图构建方法、系统及介质
CN111881827A (zh) * 2020-07-28 2020-11-03 浙江商汤科技开发有限公司 目标检测方法及装置、电子设备和存储介质

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6559535B2 (ja) * 2015-10-22 2019-08-14 株式会社東芝 障害物マップ生成装置、その方法、及び、そのプログラム
EP3252658B1 (en) * 2016-05-30 2021-08-11 Kabushiki Kaisha Toshiba Information processing apparatus and information processing method
CN109325967B (zh) * 2018-09-14 2023-04-07 腾讯科技（深圳）有限公司 目标跟踪方法、装置、介质以及设备
JP7052663B2 (ja) * 2018-09-26 2022-04-12 トヨタ自動車株式会社 物体検出装置、物体検出方法及び物体検出用コンピュータプログラム
CN109597087B (zh) * 2018-11-15 2022-07-01 天津大学 一种基于点云数据的3d目标检测方法
CN109684920B (zh) * 2018-11-19 2020-12-11 腾讯科技（深圳）有限公司 物体关键点的定位方法、图像处理方法、装置及存储介质
CN110688905B (zh) * 2019-08-30 2023-04-18 中山大学 一种基于关键帧的三维物体检测与跟踪方法
CN111308993B (zh) * 2020-02-13 2022-04-01 青岛联合创智科技有限公司 一种基于单目视觉的人体目标跟随方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109188457A (zh) * 2018-09-07 2019-01-11 百度在线网络技术（北京）有限公司 物体检测框的生成方法、装置、设备、存储介质及车辆
WO2020128650A1 (en) * 2018-12-19 2020-06-25 Sony Corporation Point cloud coding structure
CN110728210A (zh) * 2019-09-25 2020-01-24 上海交通大学 一种三维点云数据的半监督目标标注方法和系统
CN111427979A (zh) * 2020-01-15 2020-07-17 深圳市镭神智能系统有限公司 基于激光雷达的动态地图构建方法、系统及介质
CN111881827A (zh) * 2020-07-28 2020-11-03 浙江商汤科技开发有限公司 目标检测方法及装置、电子设备和存储介质

Also Published As

Publication number Publication date
TW202205139A (zh) 2022-02-01
TWI758205B (zh) 2022-03-11
CN111881827A (zh) 2020-11-03
JP2022546201A (ja) 2022-11-04
CN111881827B (zh) 2022-04-26
KR20220027202A (ko) 2022-03-07

Similar Documents

Publication Publication Date Title
WO2022021872A1 (zh) 目标检测方法及装置、电子设备和存储介质
TWI781359B (zh) 人臉和人手關聯檢測方法及裝置、電子設備和電腦可讀儲存媒體
CN110287874B (zh) 目标追踪方法及装置、电子设备和存储介质
TWI766286B (zh) 圖像處理方法及圖像處理裝置、電子設備和電腦可讀儲存媒介
WO2021051857A1 (zh) 目标对象匹配方法及装置、电子设备和存储介质
WO2020135127A1 (zh) 一种行人识别方法及装置
US11301726B2 (en) Anchor determination method and apparatus, electronic device, and storage medium
US20210319538A1 (en) Image processing method and device, electronic equipment and storage medium
US11288531B2 (en) Image processing method and apparatus, electronic device, and storage medium
CN111105454B (zh) 一种获取定位信息的方法、装置及介质
TWI767596B (zh) 場景深度和相機運動預測方法、電子設備和電腦可讀儲存介質
WO2020228418A1 (zh) 视频处理方法及装置、电子设备和存储介质
WO2020042826A1 (zh) 视频流降噪方法和装置、电子设备及存储介质
WO2022017140A1 (zh) 目标检测方法及装置、电子设备和存储介质
CN111104920A (zh) 视频处理方法及装置、电子设备和存储介质
WO2022110776A1 (zh) 定位方法及装置、电子设备、存储介质、计算机程序产品、计算机程序
CN109344703B (zh) 对象检测方法及装置、电子设备和存储介质
CN111680646B (zh) 动作检测方法及装置、电子设备和存储介质
US20220383517A1 (en) Method and device for target tracking, and storage medium
CN108171222B (zh) 一种基于多流神经网络的实时视频分类方法及装置
CN113792622A (zh) 帧率调整方法及装置、电子设备和存储介质
WO2022099988A1 (zh) 目标跟踪方法及装置、电子设备和存储介质
WO2022141969A1 (zh) 图像分割方法及装置、电子设备、存储介质和程序
CN107480773B (zh) 训练卷积神经网络模型的方法、装置及存储介质
WO2022110801A1 (zh) 数据处理方法及装置、电子设备和存储介质

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2022505272

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20227003199

Country of ref document: KR

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: KR1020227003199

Country of ref document: KR

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21850838

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21850838

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 21850838

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 28.08.2023)