WO2020228353A1 - 一种基于运动加速度的图像搜索方法、系统及电子设备 - Google Patents

一种基于运动加速度的图像搜索方法、系统及电子设备 Download PDF

Info

Publication number
WO2020228353A1
WO2020228353A1 PCT/CN2019/130538 CN2019130538W WO2020228353A1 WO 2020228353 A1 WO2020228353 A1 WO 2020228353A1 CN 2019130538 W CN2019130538 W CN 2019130538W WO 2020228353 A1 WO2020228353 A1 WO 2020228353A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
target
frame
tracked
search range
Prior art date
Application number
PCT/CN2019/130538
Other languages
English (en)
French (fr)
Inventor
张昱航
任宏帅
叶可江
王洋
须成忠
Original Assignee
深圳先进技术研究院
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 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2020228353A1 publication Critical patent/WO2020228353A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/53—Querying
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
    • G06V20/47—Detecting features for summarising video content

Definitions

  • This application belongs to the field of image search technology, and in particular relates to an image search method, system and electronic device based on motion acceleration.
  • Target tracking technology in the video has received extensive attention from universities and enterprises.
  • the general technical solution is to mark the target position to be tracked in the first frame of the video, and then in each subsequent frame, perform a global search to find the target in the next frame.
  • Target to be tracked It is usually achieved in the following ways:
  • the existing image search technologies are all global search, which will cause longer retrieval time and more computational redundancy.
  • the present application provides an image search method, system and electronic device based on motion acceleration, aiming to solve one of the above-mentioned technical problems in the prior art at least to a certain extent.
  • An image search method based on motion acceleration includes the following steps:
  • Step a Calculate the acceleration of the target to be tracked in the current frame of image according to the displacement of the previous two frames of image;
  • Step b Determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration calculation result
  • Step c Extract the candidate frame of the target to be tracked in the current frame image along the diagonal of the rectangular frame of the search range through the RPN network, and perform feature analysis on the candidate frame to obtain the target to be tracked in the current frame image In the location.
  • the acceleration is a vector unit, which has both speed and direction
  • the acceleration calculation formula is:
  • the technical solution adopted in the embodiment of the application further includes: in the step b, determining the search range rectangle of the target to be tracked in the current frame image according to the acceleration calculation result specifically includes: The center position of the tracking target is taken as the intersection of the diagonals of the search range rectangle, defining The horizontal and vertical coordinates respectively represent the center position of the target to be tracked, and define the starting search origin of the next frame, i+2 frame as:
  • the length and width of the search range rectangle are:
  • width i+2 2*width i+1
  • height i+2 2*height i+2 .
  • the technical solution adopted by the embodiment of the application further includes: in the step c, the extraction of the candidate frame of the target to be tracked in the current frame image along the diagonal of the search range rectangle through the RPN network is specifically: Obtain three points on the diagonal line of the rectangular frame of the search range according to the preset interval distance, and then scale again according to the set three aspect ratios to obtain nine candidate frames.
  • the technical solution adopted by the embodiment of the present application further includes: in the step a, the first two frames of images are specifically two consecutive frames, two frames of images at discrete intervals, or two frames of images at any time.
  • an image search system based on motion acceleration including:
  • Acceleration calculation module used to calculate the acceleration of the target to be tracked in the current frame of image according to the displacement of the previous two frames of image;
  • Search range calculation module used to determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration calculation result
  • Candidate frame extraction module used to extract the candidate frame of the target to be tracked in the current frame image along the diagonal of the search range rectangular frame through the RPN network;
  • Target retrieval module used to perform feature analysis on the candidate frame to obtain the position of the target to be tracked in the current frame of image.
  • the technical solution adopted in the embodiment of the present application further includes: the acceleration is a vector unit, which has both speed and direction, and the acceleration calculation formula is:
  • the technical solution adopted in the embodiment of the application further includes: the search range calculation module determines the search range rectangle of the target to be tracked in the current frame image according to the acceleration calculation result, specifically including: The center position is taken as the intersection of the diagonals of the search range rectangle, defining The horizontal and vertical coordinates respectively represent the center position of the target to be tracked, and define the starting search origin of the next frame, i+2 frame as:
  • the length and width of the search range rectangle are:
  • width i+2 2*width i+1
  • height i+2 2*height i+2 .
  • the technical solution adopted in the embodiment of the present application further includes: the candidate frame extraction module extracts the candidate frame of the target to be tracked in the current frame image through the RPN network along the diagonal of the search range rectangle, specifically: in the search range rectangle Three points are obtained on the diagonal line of the frame according to the preset interval distance, and then scaled again according to the set three aspect ratios to obtain nine candidate frames.
  • the technical solution adopted in the embodiment of the present application further includes: the first two frames of images are specifically two consecutive frames of images, two frames of images at discrete intervals, or two frames of images at any time.
  • an electronic device including:
  • At least one processor At least one processor
  • a memory communicatively connected with the at least one processor; wherein,
  • the memory stores instructions executable by the one processor, and the instructions are executed by the at least one processor, so that the at least one processor can perform the following operations of the above-mentioned motion acceleration-based image search method :
  • Step a Calculate the acceleration of the target to be tracked in the current frame of image according to the displacement of the previous two frames of image;
  • Step b Determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration calculation result
  • Step c Extract the candidate frame of the target to be tracked in the current frame image along the diagonal of the rectangular frame of the search range through the RPN network, and perform feature analysis on the candidate frame to obtain the target to be tracked in the current frame image In the location.
  • the beneficial effects produced by the embodiments of the present application are: the image search method, system and electronic device based on motion acceleration in the embodiments of the present application use acceleration calculation methods to determine a limited search range rectangle, and search along The candidate frame of the tracking target is selected at three points on the diagonal of the scope rectangle, thereby determining a smaller search scope.
  • this application does not need to perform a global search, which greatly reduces the search scope and reduces The amount of calculation increases the calculation speed.
  • Fig. 1 is a flowchart of an image search method based on motion acceleration according to an embodiment of the present application
  • Figure 2(a) is the target screen of the target to be tracked in the i-th frame
  • Figure 2(b) is the target screen of the target to be tracked in the (i+1)th frame
  • Figure 3 is a schematic diagram of the acceleration calculation method under the same shooting screen (the camera is fixed);
  • Figure 4 is a schematic diagram of a rectangular frame of the search range of frame i+2;
  • FIG. 5 is a schematic diagram of a generation rule of an RPN network according to an embodiment of the application.
  • FIG. 6 is a schematic structural diagram of an image search system based on motion acceleration according to an embodiment of the present application.
  • FIG. 7 is a schematic diagram of a hardware device structure of an image search method based on motion acceleration provided by an embodiment of the present application.
  • FIG. 1 is a flowchart of an image search method based on motion acceleration according to an embodiment of the present application.
  • the image search method based on motion acceleration in the embodiment of the present application includes the following steps:
  • Step 100 Mark the position of the target to be tracked in the first frame of image at the beginning of the video
  • Step 200 Calculate the acceleration of the target to be tracked in the current frame of image and its possible direction according to the displacement of the previous two frames of image;
  • Fig. 2(a) is the target picture of the target to be tracked in the i-th frame
  • Fig. 2(b) is the target picture of the target to be tracked in the (i+1)th frame.
  • the target tracking process in this figure is abstracted into a mathematical form, which can be expressed as shown in Figure 3, which is the acceleration calculation method under the same shooting frame (the camera is fixed).
  • Figure 3 is the acceleration calculation method under the same shooting frame (the camera is fixed).
  • the acceleration also maintains the same properties as the acceleration in physics, both are vector units, and have both speed and direction.
  • the specific acceleration calculation formula is:
  • the present application is not limited to determining the acceleration of the third frame based on the image displacement of the first two consecutive frames, and the acceleration calculation of the current frame may be performed using discretely spaced frames or target displacements in two frames of pictures at any time.
  • Step 300 Determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration and direction calculation results
  • the calculation method of the search range rectangle is specifically: taking the center position of the target to be tracked in the i+1th frame as the intersection of the diagonals of the search range rectangle, defining Respectively represent the horizontal and vertical coordinates of the center position of the target to be tracked, and define the starting search origin of the next frame, i+2 frame:
  • the formulas (2) and (3) determine the lower left origin position of the start search of frame i+2, and the start point of the search range rectangle box of frame i+2 is The length and width of the search range rectangle are:
  • Figure 4 is a schematic diagram of a rectangular frame of the search range of frame i+2.
  • the actual detection range of the i+2 frame has changed from the entire picture of the traditional algorithm to the rectangular frame in the box in the upper right picture (for clear display, the picture size of the i+2 frame has been enlarged, and it is actually the whole The size of the picture has not changed, only the search range rectangle), thereby reducing the amount of calculation for the target search.
  • the starting search origin of the next frame can be determined by the coordinate origin or the four boundary vertices of the previous frame.
  • Step 400 Take three points along the diagonal of the rectangular frame of the search range through the RPN network according to the set interval distance, and extract 9 candidate frames of the target to be tracked in the current frame image according to the three aspect ratios;
  • the existing detection method is to first perform the operations of unchanged original size, 0.5 scaling the original image, and 2 times expanding the original image on each of the selected center positions of the image in the entire image, and then perform operations in these three
  • the aspect ratio of the image size is 1:1, 1:2, 2:1. Therefore, 3*3 candidate frames can be found for selection at each center point position. There are many redundant candidate frames generated in this way.
  • this application no longer adopts candidate frames of three scales, that is, no longer do the original size of the original picture unchanged, 0.5 scale original picture, 2 times The operation of expanding the original picture, but using three points along the diagonal of the search range rectangle to select the candidate frame.
  • Step 500 Perform feature analysis on 9 candidate frames to obtain the position of the target to be tracked in the current frame of image.
  • FIG. 5 is a schematic diagram of a candidate frame generation rule in an embodiment of this application.
  • the nine boxes in Figure 5 are the generated candidate boxes for the target to be tracked. These nine frames are all obtained by uniformly magnifying the candidate frames of fixed size by 1.25 times and then extracting them according to three different aspect ratios.
  • D in Figure 5 is the diagonal diameter of the rectangular box of the search range. Three points are obtained on the diagonal diagonal with an interval of 0.25, 0.5, and 0.75, respectively, and then according to 1:1, 1:2 The three aspect ratios of 2:1 are scaled again to obtain nine candidate frames.
  • the existing RPN network detects and compares 9N candidate frames with N center points on all pictures. However, this application only needs to detect 9 candidate frames after determining the search range rectangle, which greatly reduces the search range. It can be understood that parameters such as the distance between points taken on the diagonal line and the zoom aspect ratio can be set according to actual operations.
  • the pixel value or feature value at this position of the previous picture is retained at this step until the selected candidate frame can accurately capture all the search range rectangles.
  • FIG. 6 is a schematic structural diagram of an image search system based on motion acceleration according to an embodiment of the present application.
  • the image search system based on motion acceleration in the embodiment of the present application includes a position marking module, an acceleration calculation module, a search range calculation module, a candidate frame extraction module, and a target retrieval module.
  • Position marking module used to mark the position of the target to be tracked in the first frame of the video
  • Acceleration calculation module used to calculate the acceleration of the target to be tracked in the current frame image and its possible direction according to the displacement of the previous two frames of images; specifically, as shown in Figure 2, where Figure 2(a) is the to be tracked The target is in the target screen of the i-th frame. Figure 2(b) is the target screen of the target to be tracked in the (i+1)th frame.
  • the target tracking process in this figure is abstracted into a mathematical form, which can be expressed as shown in Figure 3, which is the acceleration calculation method under the same shooting frame (the camera is fixed).
  • the acceleration also maintains the same properties as the acceleration in physics, both are vector units, and have both speed and direction.
  • the specific acceleration calculation formula is:
  • the present application is not limited to determining the acceleration of the third frame according to the displacement of the first two frames of the image, and the acceleration calculation can also be performed using discretely spaced frames or target displacements in two frames of pictures at any time.
  • Search range calculation module used to determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration and direction calculation results; wherein the calculation method of the search range rectangle is specifically as follows: The center position is taken as the intersection of the diagonals of the search range rectangle, defining Respectively represent the horizontal and vertical coordinates of the center position of the target to be tracked, and define the starting search origin of the next frame, i+2 frame:
  • the formulas (2) and (3) determine the lower left origin position of the start search of frame i+2, and the start point of the search range rectangle box of frame i+2 is The length and width of the rectangular frame are:
  • Figure 4 is a schematic diagram of a rectangular frame of the search range of frame i+2.
  • the actual detection range of frame i+2 has changed from the entire picture of the traditional algorithm to the search range rectangle in the box in the upper right picture (for clarity, the picture size of frame i+2 has been enlarged.
  • the size of the entire picture remains the same, only the search range rectangle), which reduces the amount of calculation for target search.
  • the starting search origin of the next frame can be determined by the coordinate origin or the four boundary vertices of the previous frame.
  • Candidate frame extraction module used to take three points along the diagonal of the search range rectangular frame through the RPN network according to the set interval distance, and extract 9 of the target to be tracked in the current frame image according to the three aspect ratios Candidate frame; among them, in order to save computing power and improve speed, this application no longer adopts three-scale candidate frames, that is, no longer perform the operations of unchanged original size of the original picture, zooming the original picture by 0.5, and expanding the original picture by 2 times. Instead, three points along the diagonal of the search range rectangle are used to select candidate frames.
  • FIG. 5 is a schematic diagram of the generation rule of the RPN network according to the embodiment of the application.
  • the nine boxes in the figure are the generated candidate boxes for the target to be tracked. These nine frames are all obtained by uniformly magnifying the candidate frames of fixed size by 1.25 times and then extracting them according to three different aspect ratios.
  • D in Figure 4 is the diagonal diameter of the rectangular box of the search range. Three points are obtained on the diagonal diagonal with an interval of 0.25, 0.5, and 0.75, respectively, and then according to 1:1, 1:2 The three aspect ratios of 2:1 are scaled again to obtain nine candidate frames.
  • the existing RPN network detects and compares 9N candidate frames with N center points on all pictures. However, this application only needs to detect 9 candidate frames after determining the search range rectangle, which greatly reduces the search range. It can be understood that parameters such as the distance between points taken on the diagonal line and the zoom aspect ratio can be set according to actual operations.
  • the pixel value or feature value at this position of the previous picture is retained at this step until the selected candidate frame can accurately capture all the search range rectangles.
  • Target retrieval module used to perform feature analysis on 9 candidate frames to obtain the position of the target to be tracked in the current frame of image.
  • FIG. 7 is a schematic diagram of a hardware device structure of an image search method based on motion acceleration provided by an embodiment of the present application.
  • the device includes one or more processors and memory. Taking a processor as an example, the device may also include: an input system and an output system.
  • the processor, the memory, the input system, and the output system may be connected by a bus or in other ways.
  • the connection by a bus is taken as an example.
  • the memory can be used to store non-transitory software programs, non-transitory computer executable programs, and modules.
  • the processor executes various functional applications and data processing of the electronic device by running non-transitory software programs, instructions, and modules stored in the memory, that is, realizing the processing methods of the foregoing method embodiments.
  • the memory may include a program storage area and a data storage area, where the program storage area can store an operating system and an application program required by at least one function; the data storage area can store data and the like.
  • the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid state storage devices.
  • the storage may optionally include storage remotely arranged with respect to the processor, and these remote storages may be connected to the processing system through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the input system can receive input digital or character information, and generate signal input.
  • the output system may include display devices such as a display screen.
  • the one or more modules are stored in the memory, and when executed by the one or more processors, the following operations of any of the foregoing method embodiments are performed:
  • Step a Calculate the acceleration of the target to be tracked in the current frame of image according to the displacement of the previous two frames of image;
  • Step b Determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration calculation result
  • Step c Extract the candidate frame of the target to be tracked in the current frame image along the diagonal of the rectangular frame of the search range through the RPN network, and perform feature analysis on the candidate frame to obtain the target to be tracked in the current frame image In the location.
  • the embodiments of the present application provide a non-transitory (non-volatile) computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can perform the following operations:
  • Step a Calculate the acceleration of the target to be tracked in the current frame of image according to the displacement of the previous two frames of image;
  • Step b Determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration calculation result
  • Step c Extract the candidate frame of the target to be tracked in the current frame image along the diagonal of the rectangular frame of the search range through the RPN network, and perform feature analysis on the candidate frame to obtain the target to be tracked in the current frame image In the location.
  • the embodiment of the present application provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer To make the computer do the following:
  • Step a Calculate the acceleration of the target to be tracked in the current frame of image according to the displacement of the previous two frames of image;
  • Step b Determine the search range rectangle of the target to be tracked in the current frame of image according to the acceleration calculation result
  • Step c Extract the candidate frame of the target to be tracked in the current frame image along the diagonal of the rectangular frame of the search range through the RPN network, and perform feature analysis on the candidate frame to obtain the target to be tracked in the current frame image In the location.
  • the image search method, system and electronic device based on motion acceleration use an acceleration calculation method to determine a rectangular frame with a limited search range, and track the target candidate frame along three points on the diagonal of the rectangular frame of the search range By selecting, a smaller search scope is determined. Compared with the prior art, this application does not need to perform a global search, which greatly reduces the search scope, reduces the amount of calculation, and improves the calculation speed.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

一种基于运动加速度的图像搜索方法、系统及电子设备。所述方法包括：步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。该方案利用加速度计算方式确定的一个有限的搜索范围矩形框，并沿搜索范围矩形框对角线进行追踪目标的候选框选定，确定了一个更小的检索范围，无需进行全局检索，大大缩小了搜索范围从而减少了计算量，提高了计算速度。

Description

一种基于运动加速度的图像搜索方法、系统及电子设备 技术领域
本申请属于图像搜索技术领域，特别涉及一种基于运动加速度的图像搜索方法、系统及电子设备。
背景技术
随着人工智能技术的发展，越来越多的前沿知识实现了落地，其中，视频中的物体(目标)追踪技术受到了高校和企业界的广泛关注。目前，对于视频中的目标追踪，一般采用的技术方案是在视频的开始的第一帧标记出待追踪目标位置，然后再接下来的每一帧中，进行全局搜索从而找到下一帧中的待追踪目标。通常采用以下几种方式实现：
一、对全局图像进行滑动窗口方式下的搜索[Girshick R B,Donahue J,Darrell T,et al.Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[J].computer vision and pattern recognition,2014:580-587.]，这种搜索方式的效率相对低下，并不能克服物体在运动过程中的形变。
二、采用区域提议网络(RPN，Region Proposal Network，区域生成网络)进行[Ren S,He K,Girshick R B,et al.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.]，这种网络的好处在于人为设置了全局目标的搜索方式，但是这种方式仍然要计算大量的图像，搜索范围广。
如上所述，现有的图像搜索技术都是全局范围搜索，都会产生较长的检索时间和比较多的计算冗余。
发明内容
本申请提供了一种基于运动加速度的图像搜索方法、系统及电子设备，旨在至少在一定程度上解决现有技术中的上述技术问题之一。
为了解决上述问题，本申请提供了如下技术方案：
一种基于运动加速度的图像搜索方法，包括以下步骤：
步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
本申请实施例采取的技术方案还包括：在所述步骤a中，所述加速度为矢量单位，既有速度也有方向，所述加速度计算公式为：
Figure PCTCN2019130538-appb-000001
本申请实施例采取的技术方案还包括：在所述步骤b中，所述根据加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框具体包括：将第i+1帧图像中待追踪目标的中心位置作为搜索范围矩形框对角线的交点，定义
Figure PCTCN2019130538-appb-000002
分别表示待追踪目标的中心位置的横、纵坐标，定义下一帧即i+2帧的起始搜索原点为：
Figure PCTCN2019130538-appb-000003
Figure PCTCN2019130538-appb-000004
则i+2帧的搜索范围矩形框的起始点为
Figure PCTCN2019130538-appb-000005
搜索范 围矩形框的长宽分别为：
width i+2＝2*width i+1,height i+2＝2*height i+2。
本申请实施例采取的技术方案还包括：在所述步骤c中，所述通过RPN网络沿搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框具体为：在所述搜索范围矩形框的斜对角线上分别按照预设的间隔距离取得三个点，然后分别按照设定的三种长宽尺度比进行再次缩放，得到九个候选框。
本申请实施例采取的技术方案还包括：所述步骤a中，所述前两帧图像具体为连续的两帧图像、离散间隔的两帧图像或任意时刻的两帧图像。
本申请实施例采取的另一技术方案为：一种基于运动加速度的图像搜索系统，包括：
加速度计算模块：用于根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
搜索范围计算模块：用于根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
候选框提取模块：用于通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框；
目标检索模块：用于对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
本申请实施例采取的技术方案还包括：所述加速度为矢量单位，既有速度也有方向，所述加速度计算公式为：
Figure PCTCN2019130538-appb-000006
本申请实施例采取的技术方案还包括：所述搜索范围计算模块根据加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框具体包括：将第 i+1帧图像中待追踪目标的中心位置作为搜索范围矩形框对角线的交点，定义
Figure PCTCN2019130538-appb-000007
分别表示待追踪目标的中心位置的横、纵坐标，定义下一帧即i+2帧的起始搜索原点为：
Figure PCTCN2019130538-appb-000008
Figure PCTCN2019130538-appb-000009
则i+2帧的搜索范围矩形框的起始点为
Figure PCTCN2019130538-appb-000010
搜索范围矩形框的长宽分别为：
width i+2＝2*width i+1,height i+2＝2*height i+2。
本申请实施例采取的技术方案还包括：所述候选框提取模块通过RPN网络沿搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框具体为：在所述搜索范围矩形框的斜对角线上分别按照预设的间隔距离取得三个点，然后分别按照设定的三种长宽尺度比进行再次缩放，得到九个候选框。
本申请实施例采取的技术方案还包括：所述前两帧图像具体为连续的两帧图像、离散间隔的两帧图像或任意时刻的两帧图像。
本申请实施例采取的又一技术方案为：一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述的基于运动加速度的图像搜索方法的以下操作：
步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索 范围矩形框；
步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
相对于现有技术，本申请实施例产生的有益效果在于：本申请实施例的基于运动加速度的图像搜索方法、系统及电子设备利用加速度计算方式确定的一个有限的搜索范围矩形框，并沿搜索范围矩形框对角线上三个点进行追踪目标的候选框选定，从而确定了一个更小的检索范围，相对于现有技术，本申请无需进行全局检索，大大缩小了搜索范围从而减少了计算量，提高了计算速度。
附图说明
图1是本申请实施例的基于运动加速度的图像搜索方法的流程图；
图2(a)为待追踪目标在第i帧的目标画面，图2(b)为待追踪目标在第(i+1)帧的目标画面；
图3为同一拍摄画面下(摄像头固定不动)的加速度计算方式示意图；
图4为i+2帧搜索范围矩形框示意图；
图5为本申请实施例的RPN网络的生成规则示意图；
图6是本申请实施例的基于运动加速度的图像搜索系统的结构示意图；
图7是本申请实施例提供的基于运动加速度的图像搜索方法的硬件设备结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
请参阅图1，是本申请实施例的基于运动加速度的图像搜索方法的流程图。本申请实施例的基于运动加速度的图像搜索方法包括以下步骤：
步骤100：在视频开始的第一帧图像中标记出待追踪目标的位置；
步骤200：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度及其可能的方向；
步骤200中，如图2所示，其中，图2(a)为待追踪目标在第i帧的目标画面，图2(b)为待追踪目标在第(i+1)帧的目标画面。将该图的目标追踪过程抽象为数学形式，即可表示为图3所示，为同一拍摄画面下(摄像头固定不动)的加速度计算方式。本申请实施例中，加速度同样和物理学中的加速度保持一样的性质，均为矢量单位，既有速度也有方向。加速度计算公式具体为：
Figure PCTCN2019130538-appb-000011
可以理解，本申请不仅限于根据连续的前两帧图像位移来确定第三帧的加速度，还可以采用离散间隔的帧或任意时刻两帧图片中的目标位移进行当前帧的加速度计算。
步骤300：根据加速度和方向计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
步骤300中，搜索范围矩形框计算方式具体为：将第i+1帧中待追踪目标的中心位置作为搜索范围矩形框对角线的交点，定义
Figure PCTCN2019130538-appb-000012
分别表示待追踪目标的中心位置的横、纵坐标，定义下一帧即i+2帧的起始搜索原点为：
Figure PCTCN2019130538-appb-000013
Figure PCTCN2019130538-appb-000014
公式(2)、(3)确定了i+2帧起始搜索的左下原点位置，则i+2帧的搜索范围矩形框的起始点为
Figure PCTCN2019130538-appb-000015
搜索范围矩形框长宽分别为：
width i+2＝2*width i+1,height i+2＝2*height i+2  (4)
将上述过程绘制成如图4所示，即为i+2帧搜索范围矩形框示意图。如图所示，i+2帧的实际检测范围由传统算法的整张图片变为了右上方图片中方框内的矩形框(为清晰显示，i+2帧的图片尺寸做了放大，实际上整张图片尺寸一直不变，变的只有搜索范围矩形框)，从而减少目标搜索的计算量。
可以理解，因为全等四边形的中心和四个点均能确定唯一的矩形框，因此无论是通过坐标原点还是通过前一帧的四个边界顶点都可以确定下一帧的起始搜索原点。
步骤400：通过RPN网络沿搜索范围矩形框的对角线按照设定的间隔距离取三个点，并分别按照三种长宽比提取到待追踪目标在当前帧图像中的9个候选框；
步骤400中，现有的检测方式为在整张图像中，先对图像每一个选出的中心位置分别进行原始尺寸不变、0.5缩放原始图片、2倍扩大原始图像的操作，继而在这三种图像的尺寸上进行长宽比分别为1:1、1:2、2:1的改变。所以每一个中心点位置可以找到3*3种候选框进行选择。该方式生成的冗余候选框比较多，为了节省计算力并且提升速度，本申请不再采用三种尺度的候选框，即不再进行原始图片的原始尺寸不变、0.5缩放原始图片、2倍扩大原始图片的操作，而是采用沿搜索范围矩形框对角线上三个点进行候选框的选定。
步骤500：对9个候选框进行特征分析，得到待追踪目标在当前帧图像中的位置。
具体请参阅图5，为本申请实施例的候选框生成规则示意图。图5中的九个框即为生成的待追踪目标的候选框。这九个框都是将固定大小的候选框统一放大1.25倍后再按照三种不同的长宽比提取得到的。图5中的D即为搜索范围矩形框的斜对角线直径长，在斜对角线上分别按照0.25、0.5、0.75的间隔距离取得三个点，然后分别按照1:1、1:2、2:1的三种长宽尺度比进行再次缩放，从而得到九个候选框。现有的RPN网络是在全部的图片上进行N个中心点的9N个候选框检测和对比，而本申请只需要在确定搜索范围矩形框后进行9个候选框的检测，极大地缩小了检索范围。可以理解，斜对角线上取点的间隔距离以及缩放长宽尺度比等参数都可以根据实际操作进行设定。
另外，如果候选框缩放后超过了搜索范围矩形框，则在这一步保留上一帧图片这个位置的像素值或者特征值，直到选出的候选框能够准确捕捉到所有搜索范围矩形框。
请参阅图6，是本申请实施例的基于运动加速度的图像搜索系统的结构示意图。本申请实施例的基于运动加速度的图像搜索系统包括位置标记模块、加速度计算模块、搜索范围计算模块、候选框提取模块和目标检索模块。
位置标记模块：用于在视频开始的第一帧图像中标记出待追踪目标的位置；
加速度计算模块：用于根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度及其可能的方向；具体的，如图2所示，其中，图2(a)为待追踪目标在第i帧的目标画面，图2(b)为待追踪目标在第(i+1)帧的目标画面。将该图的目标追踪过程抽象为数学形式，即可表示为图3所示，为同一拍摄画面下(摄像头固定不动)的加速度计算方式。本申请实施例中，加速度同样和物理学中的加速度保持一样的性质，均为矢量单位，既有速度也有方向。加速度计算公式具体为：
Figure PCTCN2019130538-appb-000016
可以理解，本申请不仅限于根据图像前两帧位移来确定第三帧的加速度，还可以采用离散间隔的帧或任意时刻两帧图片中的目标位移进行加速度计算。
搜索范围计算模块：用于根据加速度和方向计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；其中，搜索范围矩形框计算方式具体为：将第i+1帧中待追踪目标的中心位置作为搜索范围矩形框对角线的交点，定义
Figure PCTCN2019130538-appb-000017
分别表示待追踪目标的中心位置的横、纵坐标，定义下一帧即i+2帧的起始搜索原点为：
Figure PCTCN2019130538-appb-000018
Figure PCTCN2019130538-appb-000019
公式(2)、(3)确定了i+2帧起始搜索的左下原点位置，则i+2帧的搜索范围矩形框的起始点为
Figure PCTCN2019130538-appb-000020
矩形框长宽分别为：
width i+2＝2*width i+1,height i+2＝2*height i+2  (4)
将上述过程绘制成如图4所示，即为i+2帧搜索范围矩形框示意图。如图所示，i+2帧的实际检测范围由传统算法的整张图片变为了右上方图片中方框内的搜索范围矩形框(为清晰显示，i+2帧的图片尺寸做了放大，实际上整张图片尺寸一直不变，变的只有搜索范围矩形框)，从而减少目标搜索的计算量。
可以理解，因为全等四边形的中心和四个点均能确定唯一的矩形框，因此无论是通过坐标原点还是通过前一帧的四个边界顶点都可以确定下一帧的起始搜索原点。
候选框提取模块：用于通过RPN网络沿搜索范围矩形框对角线按照设定的间隔距离取三个点，并分别按照三种长宽比提取到待追踪目标在当前帧图像 中的9个候选框；其中，为了节省计算力并且提升速度，本申请不再采用三种尺度的候选框，即不再进行原始图片的原始尺寸不变、0.5缩放原始图片、2倍扩大原始图片的操作，而是采用沿搜索范围矩形框对角线上三个点进行候选框的选定。
具体请参阅图5，为本申请实施例的RPN网络的生成规则示意图。图中的九个框即为生成的待追踪目标的候选框。这九个框都是将固定大小的候选框统一放大1.25倍后再按照三种不同的长宽比提取得到的。图4中的D即为搜索范围矩形框的斜对角线直径长，在斜对角线上分别按照0.25、0.5、0.75的间隔距离取得三个点，然后分别按照1:1、1:2、2:1的三种长宽尺度比进行再次缩放，从而得到九个候选框。现有的RPN网络是在全部的图片上进行N个中心点的9N个候选框检测和对比，而本申请只需要在确定搜索范围矩形框后进行9个候选框的检测，极大地缩小了检索范围。可以理解，斜对角线上取点的间隔距离以及缩放长宽尺度比等参数都可以根据实际操作进行设定。
另外，如果候选框缩放后超过了搜索范围矩形框，则在这一步保留上一帧图片这个位置的像素值或者特征值，直到选出的候选框能够准确捕捉到所有搜索范围矩形框。
目标检索模块：用于对9个候选框进行特征分析，得到待追踪目标在当前帧图像中的位置。
图7是本申请实施例提供的基于运动加速度的图像搜索方法的硬件设备结构示意图。如图7所示，该设备包括一个或多个处理器以及存储器。以一个处理器为例，该设备还可以包括：输入系统和输出系统。
处理器、存储器、输入系统和输出系统可以通过总线或者其他方式连接，图7中以通过总线连接为例。
存储器作为一种非暂态计算机可读存储介质，可用于存储非暂态软件程序、非暂态计算机可执行程序以及模块。处理器通过运行存储在存储器中的非暂态软件程序、指令以及模块，从而执行电子设备的各种功能应用以及数据处理，即实现上述方法实施例的处理方法。
存储器可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储数据等。此外，存储器可以包括高速随机存取存储器，还可以包括非暂态存储器，例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施例中，存储器可选包括相对于处理器远程设置的存储器，这些远程存储器可以通过网络连接至处理系统。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入系统可接收输入的数字或字符信息，以及产生信号输入。输出系统可包括显示屏等显示设备。
所述一个或者多个模块存储在所述存储器中，当被所述一个或者多个处理器执行时，执行上述任一方法实施例的以下操作：
步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
上述产品可执行本申请实施例所提供的方法，具备执行方法相应的功能模 块和有益效果。未在本实施例中详尽描述的技术细节，可参见本申请实施例提供的方法。
本申请实施例提供了一种非暂态(非易失性)计算机存储介质，所述计算机存储介质存储有计算机可执行指令，该计算机可执行指令可执行以下操作：
步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
本申请实施例提供了一种计算机程序产品，所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算机程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，使所述计算机执行以下操作：
步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
本申请实施例的基于运动加速度的图像搜索方法、系统及电子设备利用加速度计算方式确定的一个有限的搜索范围矩形框，并沿搜索范围矩形框对角线 上三个点进行追踪目标的候选框选定，从而确定了一个更小的检索范围，相对于现有技术，本申请无需进行全局检索，大大缩小了搜索范围从而减少了计算量，提高了计算速度。
以上所述仅是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本发明的保护范围。

Claims (11)

  1. 一种基于运动加速度的图像搜索方法，其特征在于，包括以下步骤：
    步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
    步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
    步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
  2. 根据权利要求1所述的基于运动加速度的图像搜索方法，其特征在于，在所述步骤a中，所述加速度为矢量单位，既有速度也有方向，所述加速度计算公式为：
    Figure PCTCN2019130538-appb-100001
  3. 根据权利要求2所述的基于运动加速度的图像搜索方法，其特征在于，在所述步骤b中，所述根据加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框具体包括：将第i+1帧图像中待追踪目标的中心位置作为搜索范围矩形框对角线的交点，定义
    Figure PCTCN2019130538-appb-100002
    分别表示待追踪目标的中心位置的横、纵坐标，定义下一帧即i+2帧的起始搜索原点为：
    Figure PCTCN2019130538-appb-100003
    Figure PCTCN2019130538-appb-100004
    则i+2帧的搜索范围矩形框的起始点为
    Figure PCTCN2019130538-appb-100005
    搜索范围矩形框的长宽分别为：
    width i+2＝2*width i+1,height i+2＝2*height i+2。
  4. 根据权利要求3所述的基于运动加速度的图像搜索方法，其特征在于，在所述步骤c中，所述通过RPN网络沿搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框具体为：在所述搜索范围矩形框的斜对角线上分别按照预设的间隔距离取得三个点，然后分别按照设定的三种长宽尺度比进行再次缩放，得到九个候选框。
  5. 根据权利要求1至4任一项所述的基于运动加速度的图像搜索方法，其特征在于，所述步骤a中，所述前两帧图像具体为连续的两帧图像、离散间隔的两帧图像或任意时刻的两帧图像。
  6. 一种基于运动加速度的图像搜索系统，其特征在于，包括：
    加速度计算模块：用于根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
    搜索范围计算模块：用于根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框；
    候选框提取模块：用于通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框；
    目标检索模块：用于对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
  7. 根据权利要求6所述的基于运动加速度的图像搜索系统，其特征在于，所述加速度为矢量单位，既有速度也有方向，所述加速度计算公式为：
    Figure PCTCN2019130538-appb-100006
  8. 根据权利要求7所述的基于运动加速度的图像搜索系统，其特征在于，所述搜索范围计算模块根据加速度计算结果确定待追踪目标在当前帧图像中的搜索范围矩形框具体包括：将第i+1帧图像中待追踪目标的中心位置作为搜索 范围矩形框对角线的交点，定义
    Figure PCTCN2019130538-appb-100007
    分别表示待追踪目标的中心位置的横、纵坐标，定义下一帧即i+2帧的起始搜索原点为：
    Figure PCTCN2019130538-appb-100008
    Figure PCTCN2019130538-appb-100009
    则i+2帧的搜索范围矩形框的起始点为
    Figure PCTCN2019130538-appb-100010
    搜索范围矩形框的长宽分别为：
    width i+2＝2*width i+1,height i+2＝2*height i+2。
  9. 根据权利要求8所述的基于运动加速度的图像搜索系统，其特征在于，所述候选框提取模块通过RPN网络沿搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框具体为：在所述搜索范围矩形框的斜对角线上分别按照预设的间隔距离取得三个点，然后分别按照设定的三种长宽尺度比进行再次缩放，得到九个候选框。
  10. 根据权利要求6至9任一项所述的基于运动加速度的图像搜索系统，其特征在于，所述前两帧图像具体为连续的两帧图像、离散间隔的两帧图像或任意时刻的两帧图像。
  11. 一种电子设备，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述1至5任一项所述的基于运动加速度的图像搜索方法的以下操作：
    步骤a：根据前两帧图像的位移计算出待追踪目标在当前帧图像中的加速度；
    步骤b：根据所述加速度计算结果确定待追踪目标在当前帧图像中的搜索范 围矩形框；
    步骤c：通过RPN网络沿所述搜索范围矩形框的对角线提取待追踪目标在当前帧图像中的候选框，并对所述候选框进行特征分析，得到所述待追踪目标在当前帧图像中的位置。
PCT/CN2019/130538 2019-05-13 2019-12-31 一种基于运动加速度的图像搜索方法、系统及电子设备 WO2020228353A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910393254.7A CN110147750B (zh) 2019-05-13 2019-05-13 一种基于运动加速度的图像搜索方法、系统及电子设备
CN201910393254.7 2019-05-13

Publications (1)

Publication Number Publication Date
WO2020228353A1 true WO2020228353A1 (zh) 2020-11-19

Family

ID=67595163

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/130538 WO2020228353A1 (zh) 2019-05-13 2019-12-31 一种基于运动加速度的图像搜索方法、系统及电子设备

Country Status (2)

Country Link
CN (1) CN110147750B (zh)
WO (1) WO2020228353A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116205914A (zh) * 2023-04-28 2023-06-02 山东中胜涂料有限公司 一种防水涂料生产智能监测系统

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110147750B (zh) * 2019-05-13 2021-08-24 深圳先进技术研究院 一种基于运动加速度的图像搜索方法、系统及电子设备
CN112800811B (zh) * 2019-11-13 2023-10-13 深圳市优必选科技股份有限公司 一种色块追踪方法、装置及终端设备
CN111008305B (zh) * 2019-11-29 2023-06-23 百度在线网络技术（北京）有限公司 一种视觉搜索方法、装置以及电子设备
CN113177918B (zh) * 2021-04-28 2022-04-19 上海大学 一种无人机对电力杆塔的智能精准巡检方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105678808A (zh) * 2016-01-08 2016-06-15 浙江宇视科技有限公司 运动目标跟踪方法及装置
WO2016131300A1 (zh) * 2015-07-22 2016-08-25 中兴通讯股份有限公司 一种自适应跨摄像机多目标跟踪方法及系统
CN108363998A (zh) * 2018-03-21 2018-08-03 北京迈格威科技有限公司 一种对象的检测方法、装置、系统和电子设备
CN110147750A (zh) * 2019-05-13 2019-08-20 深圳先进技术研究院 一种基于运动加速度的图像搜索方法、系统及电子设备

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102915552B (zh) * 2012-09-18 2014-10-08 中国科学院计算技术研究所 一种可控的火焰动画生成方法及其系统
CN104933064B (zh) * 2014-03-19 2018-02-23 株式会社理光 预测目标对象的运动参数的方法和装置
CN107346538A (zh) * 2016-05-06 2017-11-14 株式会社理光 对象跟踪方法及设备
CN106403976A (zh) * 2016-08-30 2017-02-15 哈尔滨航天恒星数据系统科技有限公司 一种基于阻抗匹配的Dijkstra最优交通路径规划方法及系统
CN106604035B (zh) * 2017-01-22 2019-10-18 北京君泊网络科技有限责任公司 一种用于视频编码和压缩的运动估计的方法
CN106877237B (zh) * 2017-03-16 2018-11-30 天津大学 一种基于航拍图像的检测输电线路中绝缘子缺失的方法
CN107180435B (zh) * 2017-05-09 2020-05-26 杭州电子科技大学 一种适用于深度图像的人体目标跟踪方法
CN109559330B (zh) * 2017-09-25 2021-09-10 北京金山云网络技术有限公司 运动目标的视觉跟踪方法、装置、电子设备及存储介质
CN109635821A (zh) * 2018-12-04 2019-04-16 北京字节跳动网络技术有限公司 图像区域的特征提取方法、装置、设备及可读介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016131300A1 (zh) * 2015-07-22 2016-08-25 中兴通讯股份有限公司 一种自适应跨摄像机多目标跟踪方法及系统
CN105678808A (zh) * 2016-01-08 2016-06-15 浙江宇视科技有限公司 运动目标跟踪方法及装置
CN108363998A (zh) * 2018-03-21 2018-08-03 北京迈格威科技有限公司 一种对象的检测方法、装置、系统和电子设备
CN110147750A (zh) * 2019-05-13 2019-08-20 深圳先进技术研究院 一种基于运动加速度的图像搜索方法、系统及电子设备

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116205914A (zh) * 2023-04-28 2023-06-02 山东中胜涂料有限公司 一种防水涂料生产智能监测系统

Also Published As

Publication number Publication date
CN110147750A (zh) 2019-08-20
CN110147750B (zh) 2021-08-24

Similar Documents

Publication Publication Date Title
WO2020228353A1 (zh) 一种基于运动加速度的图像搜索方法、系统及电子设备
WO2020238560A1 (zh) 视频目标跟踪方法、装置、计算机设备及存储介质
US11062123B2 (en) Method, terminal, and storage medium for tracking facial critical area
US11643076B2 (en) Forward collision control method and apparatus, electronic device, program, and medium
WO2018177379A1 (zh) 手势识别、控制及神经网络训练方法、装置及电子设备
US20210191611A1 (en) Method and apparatus for controlling electronic device based on gesture
CN111047626B (zh) 目标跟踪方法、装置、电子设备及存储介质
US20220282993A1 (en) Map fusion method, device and storage medium
CN110807410B (zh) 关键点定位方法、装置、电子设备和存储介质
JP7273129B2 (ja) 車線検出方法、装置、電子機器、記憶媒体及び車両
US20200410688A1 (en) Image Segmentation Method, Image Segmentation Apparatus, Image Segmentation Device
CN103440667A (zh) 一种遮挡状态下运动目标稳定追踪的自动装置
US11682212B2 (en) Hierarchical data organization for dense optical flow processing in a computer vision system
KR20220153667A (ko) 특징 추출 방법, 장치, 전자 기기, 저장 매체 및 컴퓨터 프로그램
JP2023530796A (ja) 認識モデルトレーニング方法、認識方法、装置、電子デバイス、記憶媒体及びコンピュータプログラム
CN111914756A (zh) 一种视频数据处理方法和装置
CN108961385A (zh) 一种slam构图方法及装置
CN111723713A (zh) 一种基于光流法的视频关键帧提取方法及系统
CN112183431A (zh) 实时行人数量统计方法、装置、相机和服务器
Yang et al. A light CNN based method for hand detection and orientation estimation
Wang et al. An improved YOLOv3 object detection network for mobile augmented reality
TW202123077A (zh) 物件偵測方法以及電子裝置
TWI712993B (zh) 應用於半特徵視覺式同步定位與建圖方法的半特徵視覺里程計
Liu et al. Lightweight Face Detection Algorithm under Occlusion Based on Improved CenterNet
CN114882403B (zh) 基于渐进注意力超图的视频时空动作定位方法

Legal Events

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

Ref document number: 19928387

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: 19928387

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19928387

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 14.06.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19928387

Country of ref document: EP

Kind code of ref document: A1