WO2020098076A1 - 跟踪目标的定位方法、装置、设备及存储介质 - Google Patents
跟踪目标的定位方法、装置、设备及存储介质 Download PDFInfo
- Publication number
- WO2020098076A1 WO2020098076A1 PCT/CN2018/123080 CN2018123080W WO2020098076A1 WO 2020098076 A1 WO2020098076 A1 WO 2020098076A1 CN 2018123080 W CN2018123080 W CN 2018123080W WO 2020098076 A1 WO2020098076 A1 WO 2020098076A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- target
- position information
- original
- tracking
- point
- Prior art date
- Legal status (The legal status 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 status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
- G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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 OR CALCULATING; 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 OR CALCULATING; 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20101—Interactive definition of point of interest, landmark or seed
Definitions
- the present application relates to the field of image processing technology, and in particular, to a tracking target positioning method, device, equipment, and storage medium.
- Positioning long-distance tracking targets has always been one of the most challenging tasks in video surveillance.
- some are by writing the coordinate information of the target to be tracked in the image to a file, and the tracking is started.
- the initial target information is obtained by reading the file.
- the operation cycle of this method is long, and the efficiency of obtaining the initial target information is not high, or the target to be tracked is drawn in the image by manually drawing the picture by pausing the video screen.
- Position, and after tracking is started, the tracking target is initialized based on the manually drawn position information.
- a positioning method for tracking targets including:
- the original position information of the original target point selected by the user included in the positioning request is obtained, where the original position information includes the original target point in the currently playing video data The current frame image and the original coordinates of the original target point in the current frame image;
- the target prediction model is a convolutional neural network model including two convolutional layers and two fully connected layers,
- the target prediction result includes N target areas and target position information corresponding to each of the target areas, N is a positive integer, and each of the target areas contains a target to be tracked;
- a positioning device for tracking targets including:
- the request receiving module is configured to obtain the original position information of the original target point selected by the user contained in the positioning request if the positioning request tracking the target is received, wherein the original position information includes the original target point The current frame image in the currently playing video data and the original coordinates of the original target point in the current frame image;
- the target prediction module is used to perform target prediction on the current frame image according to a preset target prediction model to obtain a target prediction result, wherein the target prediction model is composed of two convolutional layers and two fully connected layers Convolutional neural network model, the target prediction result includes N target areas and target position information corresponding to each of the target areas, N is a positive integer, and each of the target areas contains a target to be tracked;
- a distance calculation module configured to calculate the Euclidean distance between each target to be tracked and the original target point according to the target position information and the original coordinates of each target area, to obtain N distances;
- the target confirmation module is used to select the distance with the smallest value from the N distances as the target distance, acquire the target position information corresponding to the target distance, and the target to be tracked in the target area corresponding to the acquired target position information The tracking target corresponding to the original target point is determined.
- a computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, and the processor implements the computer-readable instructions to implement the positioning of the tracking target method.
- One or more non-volatile readable storage media storing computer-readable instructions, when the computer-readable instructions are executed by one or more processors, so that the one or more processors execute the above tracking when executed Target positioning method.
- FIG. 1 is a schematic diagram of an application environment of a tracking target positioning method in an embodiment of the present application
- FIG. 2 is a flowchart of a method for locating a tracking target in an embodiment of the present application
- FIG. 3 is a flowchart of step S3 in the method for locating a tracking target according to an embodiment of the present application
- FIG. 4 is a flowchart of building a target prediction model in a method for tracking target positioning in an embodiment of the present application
- FIG. 5 is an implementation flowchart of a structure for constructing a convolutional neural network target detection algorithm in a tracking target positioning method according to an embodiment of the present application
- FIG. 6 is an implementation flowchart of confirming the tracking target in the tracking target positioning method in an embodiment of the present application
- FIG. 7 is a schematic diagram of a positioning device for tracking targets in an embodiment of the present application.
- FIG. 8 is a schematic diagram of a computer device in an embodiment of the present application.
- FIG. 1 shows an application environment provided by an embodiment of the present application.
- the application environment includes a server and a client.
- the server and the client are connected through a network.
- the client is used to detect a user ’s touch operation.
- the touch operation generates a positioning request, and sends the positioning request to the server.
- the client can specifically but not limited to various personal computers, laptops, smartphones, tablets, and portable wearable devices; the server is used to process video data, services
- the terminal can be realized by an independent server or a server cluster composed of multiple servers.
- the positioning method for tracking targets provided in the embodiments of the present application is applied to the server.
- FIG. 2 illustrates an implementation process of the tracking target positioning method provided by this embodiment.
- the details are as follows:
- the original target point is a point on the location area where the target selected by the user in the video picture being played is located.
- the client detects the user's touch operation, click operation or other operations, generates a tracking target positioning request, and sends the positioning request to the server for processing.
- the touch operation refers to an operation realized by touching the screen with a human hand on the display device.
- the display device can locate the selection information input according to the icon or menu position touched by the human hand, or detect the touched position according to the position touched by the human hand coordinate.
- the click operation refers to an operation realized by a cursor click on the display device.
- the display device can locate the selection information input according to the icon or menu position clicked by the cursor, or detect the coordinates of the cursor click according to the position clicked by the cursor.
- the display device used to obtain the original position information of the original target point in the video data may specifically be a touch screen or other devices, and no specific limitation is made here.
- the current frame image refers to the currently playing picture where the original target point is located, and can be locked by the components in the display device.
- the display device of the touch screen is used in this embodiment to obtain the original position information of the original target point in the video data
- the touch screen is composed of a touch detection component and a touch screen controller, that is, the touch detection component and the touch screen controller Installed in the display device, the touch detection part is used to detect the position touched by the user and lock the current frame image of the currently playing video data, generate touch information, and then send the touch information to the touch screen controller, and the main touch screen controller
- the function is to receive touch information from the touch point detection device, and convert the position touched by the user in the touch information into contact coordinates, which are the original coordinates of the original target point.
- a touch operation can be performed on the display device playing the video
- the client detects the user's touch operation on the display device
- You can determine the current frame image of the currently playing video data and the original coordinates of the original target point selected by the user in the current frame image according to the touch operation, and generate a positioning request containing the current frame image and the original coordinates.
- the server sends the positioning request to the server.
- the server receives the positioning request and reads the original position information of the original target point carried in the positioning request.
- target prediction is performed on the current frame image to obtain the target prediction result.
- the target prediction model is a convolutional neural network model including two convolutional layers and two fully connected layers.
- the target prediction The result includes N target areas and target position information corresponding to each target area, N is a positive integer, and each target area contains a target to be tracked.
- the preset target prediction model is obtained by training according to actual business needs, and is used to predict the candidate target area in the current frame image where the original target point is located, where the candidate target area is the current frame image Contains the area of movable objects.
- Each movable object corresponds to a target area.
- the target area is displayed in the form of a rectangular frame.
- the movable object in the target area is the target to be tracked.
- the target position information of the target area includes the area The upper left corner of the coordinates, the width and height of the area.
- the convolutional layer of the target prediction model is used to extract image features of the current frame image.
- the fully connected layer outputs the target area corresponding to the image feature and the target position information corresponding to the target area according to the extracted image features.
- the current frame image is input into a preset target prediction model, the image features of the current frame image are extracted through the convolution layer in the model, and then the image features are passed through the model's fully connected layer to output the target corresponding to the image features The area and the target position information corresponding to the target area are used as the target prediction result.
- a frame of image into a preset target prediction model, extract image features of this frame of image through the convolutional layer in the model, and then pass the extracted image features through the fully connected layer of the model, which will not belong to the image
- the movable objects in the background are displayed in the form of a rectangular frame. Assuming that 6 movable objects are predicted, the target prediction result is 6 target regions and the target position information corresponding to the 6 target regions.
- the target prediction of the current frame image through the trained target prediction model, and the image feature extraction of the current frame image through the convolutional layer in the target prediction model can accurately distinguish the movable objects in the current frame image And the background, to avoid the situation that the position information of the predicted target manually drawn is inaccurate, and the predicted target is easily blocked by the background, thereby improving the accuracy of forecasting the tracking target.
- S3 Calculate the Euclidean distance between each target to be tracked and the original target point according to the target position information of each target area and the original coordinates, and obtain N distances.
- Euclidean distance is a commonly used distance definition, it can be used to express the true distance between multiple points in m-dimensional space, or the natural length of the vector, and the two points in two-dimensional space.
- the Euclidean distance between the two points is the actual distance between two points. Therefore, by calculating the Euclidean distance between each target to be tracked and the original target point, each target to be tracked and the original target can be reflected most quickly and intuitively The positional relationship between points.
- the target position information of each target area obtained in step S2 and the original coordinates of the original target point obtained in step S1 are substituted into the Euclidean distance formula. After calculation, each target to be tracked in the target area and N Euclidean distances between the original target points.
- step S2 if six target areas and target position information corresponding to the six target areas are predicted through the target prediction model, the target position information and original coordinates of the six target areas are substituted into Europe
- the Euclidean distance between the target to be tracked and the original target point can be calculated to obtain 6 distances.
- S4 Select the distance with the smallest value from the N distances as the target distance, acquire the target position information corresponding to the target distance, and determine the target to be tracked in the target area corresponding to the acquired target position information as the tracking corresponding to the original target point aims.
- N Euclidean distances are calculated in step S3, it can intuitively reflect the positional relationship between each target to be tracked and the original target point, so under normal circumstances, it can be understood as to be tracked If the position of the target is closest to the original target point, the target to be tracked is the tracking target corresponding to the original target point.
- the target distance is the distance between the original target point position and the closest target to be tracked.
- any two of the N distances obtained in step S3 are compared, the distance with the smaller value is compared, and then the distance with the smaller value is compared with other distances, and so on, until the N distances are compared
- the distance with the smallest value use the distance with the smallest value as the target distance, and use the target to be tracked in the target area corresponding to the distance with the smallest value as the tracking target corresponding to the original target point.
- step S3 continue to use the example in step S3, assuming that the calculated six distances are 45.63, 56.03, 78.24, 34.33, 40.55, and 63.06, and compare any two of the six distances, such as 45.63 and 56.03 , If 45.63 is less than 56.03, 45.63 is selected, and then 45.63 is compared with other distances. If 45.63 is compared with 40.55 and 45.63 is greater than 40.55, then 40.55 is selected, and so on, until the distance with the smallest value is obtained by comparison Is 34.33, then the target to be tracked in the target area corresponding to the distance 34.33 is used as the tracking target corresponding to the original target point.
- the target to be tracked in the target area corresponding to the distance 34.33 is used as the tracking target corresponding to the original target point.
- the image predicts the target and obtains the target prediction result, which can accurately distinguish the target and the background in the current frame image, to avoid the inaccurate location information of the artificially drawn predicted target, and the predicted target is easily blocked by the background, thereby improving the tracking target The accuracy of prediction.
- the Euclidean distance between each target to be tracked and the original target point is calculated according to the target position information of each target area and the original coordinates, and then the distance with the smallest value is selected as the target distance to obtain the target
- the target position information corresponding to the distance, and the target to be tracked in the target area corresponding to the acquired target position information is determined as the tracking target corresponding to the original target point, which ensures that the predicted target area in the current frame image is more accurate
- calculate the minimum distance between the target to be tracked and the original target point to determine the tracking target method which can ensure the accuracy of positioning the tracking target.
- This method has a small amount of calculation and is easy to operate, and the user does not need to pause The video being played shortens the operation cycle, thereby improving the positioning efficiency of the tracking target.
- the target position information corresponding to the target area includes the coordinates of the upper left corner of the target area, the width and height of the target area, that is, according to the target position information of each target area and the original
- the coordinates calculate the Euclidean distance between each target to be tracked and the original target point, and obtaining N distances includes the following steps:
- x i is the horizontal coordinate of the upper left corner of the target area corresponding to the i-th target position information
- y i is the vertical coordinate of the upper left corner of the coordinate
- w i is the i
- the width and h i of the target area corresponding to the target position information are the height of the target area corresponding to the ith target position information.
- the position information of the target area includes the coordinates of the upper left corner of the rectangular frame, the width and height of the rectangular frame, in order to accurately reflect the to-be-tracked in each target area
- the position information of the target area needs to be substituted into the center point coordinate formula to calculate the center point coordinate of each target area. Therefore, N target areas are predicted on the current frame image, according to the center
- the point coordinate formula calculates the center point coordinates of the first to Nth target areas respectively to obtain N center point coordinates, which is convenient for quickly calculating each target to be tracked based on the obtained center point coordinates of the N target areas Euclidean distance from the original target point.
- the position information of the target area is substituted into the central coordinate formula Perform calculations to obtain the coordinates of the center point of the target area as (70, 80).
- step S301 the coordinates of each center point and the original coordinates obtained in step S301 are substituted into the Euclidean distance formula for calculation, and the Euclidean between the target to be tracked and the original target point in each target area can be quickly calculated distance.
- the center point coordinate of a target area obtained is (90, 80), assuming the origin coordinate is (110, 130), the center point coordinate and the origin coordinate are substituted into the Euclidean distance formula Perform calculations to find that the Euclidean distance d 1 is approximately equal to 53.85. If the other two center point coordinates are (60, 90) and (100, 70), the two center point coordinates and origin coordinates are substituted into Europe respectively.
- the Euclidean distance d 2 corresponding to the center point coordinate (60, 90) is approximately 64.03
- the Euclidean distance d 3 corresponding to the center point coordinate (100, 70) is approximately 60.83.
- the central point coordinate formula is used to calculate the central point coordinates of the target area, and then each of the obtained central point coordinates and origin coordinates are substituted into Europe
- the Euclidean distance formula can quickly calculate the Euclidean distance between the target to be tracked and the original target point in each target area.
- the tracking target positioning method further includes the following steps:
- the sample training image data includes the MsCoCo data set and the CalTech data set.
- the MsCoCo data set is a public data set shared by the network.
- the data set includes 328,000 images, 2.5 million tags, and images containing 91 types of targets.
- This data set aims at scene understanding (scene understanding).
- the target is calibrated by precise segmentation and intercepted from complex daily scenes.
- the Caltech data set is a public data set shared by the network. It mainly includes 2 types.
- One type is Caltech-101, which contains 101 types of images. Each type has about 40 to 800 images, and most of them are 50 images / type. Each image The size is about 300x200; the other is Caltech-256, which is similar to Caltech-101 and contains 30607 images.
- the method of acquiring the sample training image data may specifically include, but not limited to, receiving the sample training image data uploaded by the user, or automatically acquiring the sample training image data from a third-party video image database, etc., or other acquisition methods.
- the third-party video image database may specifically be a database of a network video platform.
- S6 Construct the structure of the target detection algorithm of the convolutional neural network based on the deep learning framework.
- the deep learning framework includes caffe or caffe2, etc.
- caffe2 is a lightweight, modular and extensible framework, including storage interface (blob), layer structure (layer) and structure link (net) .
- blob is the standard array structure and unified storage interface of the entire framework.
- the layer serves as the basis of the layer structure for modeling and calculation, and the net serves as the collection and link of the layer.
- the details of blobs describe how information is stored and exchanged between layers and nets.
- Caffe2 specifically defines the network layer by layer using its own modeling method, that is, the network is defined from the input data to the loss layer to define the entire model from the bottom up. Information such as data and partial derivatives flow forward and backward in the network, and caffe2 uses blobs to store, exchange, and manipulate this information.
- this embodiment adopts the structure of a convolutional neural network target detection algorithm (Faster-RCNN) based on the caffe2 framework.
- the structure of the convolutional neural network target detection algorithm can be carried out by the layer in the caffe2 framework for each layer structure Define, and then link each layer structure defined through the structure link (net). Secondly, store the linked layer structure through the blob in the caffe2 framework to obtain the structure of the built Faster-RCNN.
- the structure of the Faster-RCNN constructed by the framework has a clear logical structure and clear parameter definitions. It can quickly define the position of the parameters in the structure according to the parameters, which is convenient for the modification and improvement of the structure of the Faster-RCNN.
- the structure of the constructed convolutional neural network target detection algorithm is composed of two convolutional layers and two fully connected layers.
- the first layer is the first convolution layer, respectively
- the second layer is the second convolutional layer
- the third layer is the first fully connected layer
- the fourth layer is the second fully connected layer.
- the first convolution layer can be used to convolve the input image data and perform preliminary feature extraction on the image data.
- the convolution kernel window size of the convolution layer can be set to 11 * 11, adjacent local
- the center distance of the receiving domain is set to 4, and 96 feature maps corresponding to the input image data can be extracted.
- the first convolution layer is connected with a downsampling layer, and each feature map can be downgraded through a downsampling layer.
- the core window size of the down-sampling layer is usually set to 3 * 3, the center distance of the adjacent local acceptance domain is 2, and finally, the down-sampling layer can be reduced
- the dimensioned feature map is input to the second convolutional layer.
- the second convolutional layer can be used to convolve the feature map obtained from the first convolutional layer. It is to perform further feature extraction on the feature map obtained from the first convolutional layer, improve the feature extraction accuracy, and condense the second
- the size of the convolution kernel window of the layer is set to 5 * 5, the center distance of the adjacent local acceptance domain is set to 2, and 256 feature maps can be output.
- the second convolution layer is connected to a downsampling layer, and each feature map is passed through A down-sampling layer performs dimensionality reduction to compress the feature map to ensure the integrity and accuracy of the extracted features.
- the core window size of the down-sampling layer is set to 3 * 3, and the center distance between adjacent local acceptance domains is 2.
- the dimensionality-reduced feature map obtained by the down-sampling layer is input to the first fully connected layer.
- the third layer is the first fully connected layer.
- This layer has 1000 neurons, which can be used to transform the feature map obtained in the second convolutional layer into an n-dimensional column vector.
- the weight matrix and offset of the connection layer are nonlinearly transformed to obtain a 1000-dimensional column vector and input to the second fully connected layer;
- the fourth layer is the second fully connected layer, which has m neurons, and can be used to nonlinearly divide the 1000-dimensional column vector obtained in the first fully connected layer with the weight matrix and offset of the second fully connected layer Transform to obtain an m-dimensional feature vector, where the value of m is based on the number of targets on the image, which can be determined by updating the weight matrix and offset of the second fully connected layer, that is, m targets are correspondingly set with m Neurons.
- the sample training image data is used to train the structure of the target detection algorithm of the convolutional neural network to obtain the target prediction model.
- the sample training image data is used to train the structure of the convolutional neural network target detection algorithm, which is to use the shared feature alternate training method on the MsCoCo dataset and the CalTech dataset, that is, the same in the two datasets.
- the target prediction model consists of two files, one file is the structure of the convolutional neural network target detection algorithm, and the other file is the weight and bias of each layer in the structure of the convolutional neural network target detection algorithm Bias, the weight is the weight value in the weight matrix, and the offset is the offset corresponding to the weight matrix.
- the training process is to extract the features of the sample training image data in the structure of the convolutional neural network target detection algorithm, and through the traditional stochastic gradient descent algorithm, each layer of the structure of the convolutional neural network target detection algorithm In the cost function included in the layer structure, the cost function value corresponding to each layer is calculated according to the direction of the layer structure from the front to the back.
- This cost function value can be used for the subsequent calculation of the error sensitivity in each layer, and then, through the traditional inversion Propagation algorithm, each layer of the structure of the convolutional neural network target detection algorithm, according to the direction of the layer structure from back to front, calculate the error sensitivity in each layer, and finally, the calculated error sensitivity of each layer
- the included weight value and offset are used to update the original weight value and offset in the layer structure to obtain the target prediction model.
- the structure of the target detection algorithm of the convolutional neural network is constructed based on the caffe2 framework.
- the slideable convolution kernel window is used to convolve the input sample training image data to realize the feature extraction of the sample training image data, which can ensure the feature extraction of the sample training image data Completeness, and in the process of extracting image features, it has a high degree of invariance to geometric transformations such as image deformation and translation, which can ensure the integrity of the target features of the image predicted by the trained target prediction model, and can ensure the subsequent The accuracy of identifying target information in the image.
- step S6 the structure of constructing a convolutional neural network target detection algorithm based on the deep learning framework specifically includes the following steps:
- S601 Obtain the definition information of the structure of the target detection algorithm of the convolutional neural network.
- the configuration parameters required by different network structures may be different, and the structure definition of the network structure may also be different.
- the parameter names included in the layer structure corresponding to each network structure are preset And the specific parameter data corresponding to the parameter name, so that the parameter data required for the configuration of the network structure to be constructed can be quickly obtained from the definition information according to the parameter name of the network structure to be constructed, so that the subsequent process of building the network structure can be filled quickly Into the layer structure.
- the parameter name in the layer structure is the name of the layer, the type of the layer or the definition of the type of output data, etc.
- the specific parameter data corresponding to the parameter name such as the specific parameter corresponding to the "layer name” is "conv2 (second convolutional layer ) "," Layer type “corresponds to the specific parameter” convolution (convolution) "or” output data type definition "corresponds to the specific parameter” data (data) "and so on.
- S602 Fill in the definition information into the layer structure of the corresponding convolutional layer and the fully connected layer in the deep learning framework according to the layer structure defined by the preset network to generate the structure of the convolutional neural network target detection algorithm.
- the preset network definition structure includes the name of the layer, the type of the layer, the layer parameters corresponding to the type of the layer, the type definition of the output data, the learning rate of the filtering in the filtering parameters and the filtering in the filtering parameters Decay rate, etc.
- a layer structure of the preset network definition structure includes the name of the layer, the layer type, the learning rate of filtering in the filtering parameters, and the attenuation rate of filtering in the filtering parameters, etc.
- the layer structure of the model definition structure may be set as:
- the specific parameter data corresponding to the same parameter name in the layer structure of the preset network definition in the definition information is added to the pre
- the structure of the convolutional neural network target detection algorithm can be quickly generated in the layer structure of the convolutional layer and the fully connected layer corresponding to the definition information.
- the operation is simple and fast. It is not easy to make mistakes, and the specific parameter data can be quickly modified through the parameter name, which is convenient to improve the structure of the target detection algorithm of the convolutional neural network.
- the tracking target positioning method further includes the following steps:
- the tracking target in order to further improve the accuracy of positioning the tracking target, it is possible to increase the target tracking by setting a distance threshold and sending the target to be tracked in the target area corresponding to the distance smaller than the distance threshold to the user for confirmation The accuracy of positioning.
- S9 Receive the confirmation result returned by the user, and according to the confirmation result, use the tracking target confirmed by the user as the tracking target corresponding to the original target point.
- the user when the user receives the target to be tracked in the target area corresponding to the distance less than the preset threshold obtained in step S8, the user can select a tracking target from these target to be tracked according to the original target point of the current frame image , As a confirmation result and return, in order to accurately locate the tracking target.
- the target to be tracked in the target area corresponding to the distance smaller than the threshold is sent to the user for confirmation, so that the user selects one of the target to be tracked according to the original target point of the current frame image
- the target as a confirmation result and returned, can accurately determine the positioning of the tracking target, thereby improving the accuracy of positioning the tracking target.
- a tracking target positioning device is provided, and the tracking target positioning device corresponds to the tracking target positioning method in the above embodiment in one-to-one correspondence.
- the tracking target positioning device includes a request receiving module 701, a target prediction module 702, a distance calculation module 703, and a target confirmation module 704.
- the detailed description of each functional module is as follows:
- the request receiving module 701 is used to obtain the original position information of the original target point selected by the user contained in the positioning request if the positioning request of the tracking target is received, wherein the original position information includes the currently playing video data of the original target point The original coordinates of the current frame image and the original target point in the current frame image;
- the target prediction module 702 is used to perform target prediction on the current frame image according to a preset target prediction model to obtain a target prediction result, wherein the target prediction model is a convolutional nerve including two convolutional layers and two fully connected layers Network model, target prediction results include N target areas and target position information corresponding to each target area, N is a positive integer, and each target area contains a target to be tracked;
- the target prediction model is a convolutional nerve including two convolutional layers and two fully connected layers Network model
- target prediction results include N target areas and target position information corresponding to each target area, N is a positive integer, and each target area contains a target to be tracked;
- the distance calculation module 703 is used to calculate the Euclidean distance between each target to be tracked and the original target point according to the target position information and the original coordinates of each target area to obtain N distances;
- the target confirmation module 704 is used to select the distance with the smallest value from the N distances as the target distance, acquire the target position information corresponding to the target distance, and determine the target to be tracked in the target area corresponding to the acquired target position information as the original The tracking target corresponding to the target point.
- the distance calculation module 703 includes:
- the coordinate calculation unit 7031 is used to calculate the center point coordinates of the target area corresponding to the i-th target position information according to the center point coordinate formula as follows
- x i is the horizontal coordinate of the upper left corner of the target area corresponding to the i-th target position information
- y i is the vertical coordinate of the upper left corner of the coordinate
- w i is the i
- the width of the target area corresponding to the target position information and h i are the height of the target area corresponding to the ith target position information;
- the distance calculation unit 7032 is used to calculate the coordinate of the i-th center point for each center point coordinate Euclidean distance d i from the original coordinates ( ⁇ x, ⁇ y):
- the tracking target positioning device further includes:
- the data acquisition module 705 is used to acquire sample training image data
- the model training module 707 is used to train the structure of the convolutional neural network target detection algorithm using sample training image data in the structure of the convolutional neural network target detection algorithm to obtain a target prediction model.
- the structure building module 706 includes:
- Definition obtaining unit 7061 used to obtain definition information of the structure of the convolutional neural network target detection algorithm
- the structure generating unit 7062 is used to fill the definition information into the layer structure of the corresponding convolutional layer and the fully connected layer in the deep learning framework according to the layer structure defined by the preset network to generate the target detection algorithm of the convolutional neural network. structure.
- the tracking target positioning device further includes:
- the target feedback module 708 is used to feed back the target to be tracked in the target area corresponding to the target position information of the distances of the N distances that are less than the preset threshold to the user for confirmation;
- the result receiving module 709 is used to receive the confirmation result returned by the user, and according to the confirmation result, use the tracking target confirmed by the user as the tracking target corresponding to the original target point.
- Each module in the above positioning device for tracking a target may be implemented in whole or in part by software, hardware, or a combination thereof.
- the above modules may be embedded in the hardware or independent of the processor in the computer device, or may be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
- a computer device is provided.
- the computer device may be a server, and its internal structure may be as shown in FIG. 8.
- the computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities.
- the memory of the computer device includes a non-volatile storage medium and an internal memory.
- the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
- the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium.
- the database of the computer device is used to store video data.
- the network interface of the computer device is used to communicate with external terminals through a network connection. When the computer readable instructions are executed by the processor, a positioning method for tracking the target is realized.
- a computer device which includes a memory, a processor, and computer-readable instructions stored on the memory and executable on the processor.
- the processor executes the computer-readable instructions
- the tracking target of the foregoing embodiment is implemented
- the steps of the localization method are, for example, steps S1 to S4 shown in FIG. 2.
- the processor executes the computer-readable instructions
- the functions of each module / unit of the positioning device for tracking the target in the above embodiments are implemented, for example, the functions of modules 701 to 704 shown in FIG. 7. To avoid repetition, I will not repeat them here.
- a non-volatile storage medium on which computer readable instructions are stored, and when the computer readable instructions are executed by a processor, the method for locating a tracking target in the above method embodiment is implemented, or When the computer-readable instructions are executed by the processor, the functions of each module / unit in the positioning device of the tracking target in the foregoing device embodiments are realized. To avoid repetition, I will not repeat them here.
- Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
- Volatile memory can include random access memory (RAM) or external cache memory.
- RAM random access memory
- DRAM dynamic RAM
- SDRAM synchronous DRAM
- DDRSDRAM double data rate SDRAM
- ESDRAM enhanced SDRAM
- SLDRAM synchronous chain (Synchlink)
- SLDRAM memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRAM), and memory bus dynamic RAM (RDRAM), etc.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims (20)
- 一种跟踪目标的定位方法,其特征在于,所述跟踪目标的定位方法包括:若接收到跟踪目标的定位请求,则获取所述定位请求中包含的用户选定的原始目标点的原始位置信息,其中,所述原始位置信息包括所述原始目标点在当前播放的视频数据中的当前帧图像和所述原始目标点在所述当前帧图像中的原始坐标;根据预设的目标预测模型,对所述当前帧图像进行目标预测,得到目标预测结果,其中,所述目标预测模型为包括两层卷积层和两层全连接层的卷积神经网络模型,所述目标预测结果包括N个目标区域和每个所述目标区域对应的目标位置信息,N为正整数,每个所述目标区域包含一个待跟踪目标;根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离;从所述N个距离中选取数值最小的距离作为目标距离,获取所述目标距离对应的目标位置信息,并将获取到的目标位置信息对应的目标区域中的待跟踪目标确定为所述原始目标点对应的跟踪目标。
- 如权利要求1所述的跟踪目标的定位方法,其特征在于,所述目标区域对应的目标位置信息包括目标区域的左上角坐标、所述目标区域的宽和高,所述根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离包括:其中,i为正整数,且0<i≤N,x i为第i个所述目标位置信息对应的所述目标区域的左上角坐标的横坐标,y i为所述左上角坐标的纵坐标,w i为所述第i个所述目标位置信息对应的目标区域的宽和h i为所述第i个所述目标位置信息对应的目标区域的高;
- 如权利要求1所述的跟踪目标的定位方法,其特征在于,所述根据预设的目标预测模型,对所述当前帧图像进行目标预测,得到目标预测结果之前,所述跟踪目标的定位方法还包括:获取样本训练图像数据;基于深度学习框架构建卷积神经网络目标检测算法的结构;在所述卷积神经网络目标检测算法的结构中,使用所述样本训练图像数据对所述卷积神经网络目标检测算法的结构进行训练,得到所述目标预测模型。
- 如权利要求3所述的跟踪目标的定位方法,其特征在于,所述基于深度学习框架构 建卷积神经网络目标检测算法的结构包括:获取所述卷积神经网络目标检测算法的结构的定义信息;在所述深度学习框架中按照预设的网络定义的层结构,将所述定义信息填入对应的所述卷积层和所述全连接层的层结构中,生成所述卷积神经网络目标检测算法的结构。
- 如权利要求1所述的跟踪目标的定位方法,其特征在于,在所述根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离之后,所述跟踪目标的定位方法还包括:将所述N个距离中小于预设阈值的所述距离的目标位置信息对应的目标区域中的待跟踪目标反馈给用户进行确认;接收所述用户返回的确认结果,并根据所述确认结果,将所述用户确认的跟踪目标作为所述原始目标点对应的跟踪目标。
- 一种跟踪目标的定位装置,其特征在于,所述跟踪目标的定位装置包括:请求接收模块,用于若接收到跟踪目标的定位请求,则获取所述定位请求中包含的用户选定的原始目标点的原始位置信息,其中,所述原始位置信息包括所述原始目标点在当前播放的视频数据中的当前帧图像和所述原始目标点在所述当前帧图像中的原始坐标;目标预测模块,用于根据预设的目标预测模型,对所述当前帧图像进行目标预测,得到目标预测结果,其中,所述目标预测模型为包括两层卷积层和两层全连接层的卷积神经网络模型,所述目标预测结果包括N个目标区域和每个所述目标区域对应的目标位置信息,N为正整数,每个所述目标区域包含一个待跟踪目标;距离计算模块,用于根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离;目标确认模块,用于从所述N个距离中选取数值最小的距离作为目标距离,获取所述目标距离对应的目标位置信息,并将获取到的目标位置信息对应的目标区域中的待跟踪目标确定为所述原始目标点对应的跟踪目标。
- 如权利要求6所述的跟踪目标的定位装置,其特征在于,所述跟踪目标的定位装置还包括:数据获取模块,用于获取样本训练图像数据;结构构建模块,用于基于深度学习框架构建卷积神经网络目标检测算法的结构;模型训练模块,用于在所述卷积神经网络目标检测算法的结构中,使用所述样本训练图像数据对所述卷积神经网络目标检测算法的结构进行训练,得到所述目标预测模型。
- 如权利要求7所述的跟踪目标的定位装置,其特征在于,所述结构构建模块包括:定义获取单元,用于获取卷积神经网络目标检测算法的结构的定义信息;结构生成单元,用于在深度学习框架中按照预设的网络定义的层结构,将定义信息填入对应的卷积层和全连接层的层结构中,生成卷积神经网络目标检测算法的结构。
- 如权利要求6所述的跟踪目标的定位装置,其特征在于,所述跟踪目标的定位装置还包括:目标反馈模块,用于将N个距离中小于预设阈值的距离的目标位置信息对应的目标区域中的待跟踪目标反馈给用户进行确认;结果接收模块,用于接收用户返回的确认结果,并根据确认结果,将用户确认的跟踪目标作为原始目标点对应的跟踪目标。
- 一种计算机设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令,其特征在于,所述处理器执行所述计算机可读指令时实现如下步骤:若接收到跟踪目标的定位请求,则获取所述定位请求中包含的用户选定的原始目标点的原始位置信息,其中,所述原始位置信息包括所述原始目标点在当前播放的视频数据中的当前帧图像和所述原始目标点在所述当前帧图像中的原始坐标;根据预设的目标预测模型,对所述当前帧图像进行目标预测,得到目标预测结果,其中,所述目标预测模型为包括两层卷积层和两层全连接层的卷积神经网络模型,所述目标预测结果包括N个目标区域和每个所述目标区域对应的目标位置信息,N为正整数,每个所述目标区域包含一个待跟踪目标;根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离;从所述N个距离中选取数值最小的距离作为目标距离,获取所述目标距离对应的目标位置信息,并将获取到的目标位置信息对应的目标区域中的待跟踪目标确定为所述原始目标点对应的跟踪目标。
- 如权利要求11所述的计算机设备,其特征在于,所述目标区域对应的目标位置信息包括目标区域的左上角坐标、所述目标区域的宽和高,所述根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离包括:其中,i为正整数,且0<i≤N,x i为第i个所述目标位置信息对应的所述目标区域的左上角坐标的横坐标,y i为所述左上角坐标的纵坐标,w i为所述第i个所述目标位置信息对应的目标区域的宽和h i为所述第i个所述目标位置信息对应的目标区域的高;
- 如权利要求11所述的计算机设备,其特征在于,所述根据预设的目标预测模型,对所述当前帧图像进行目标预测,得到目标预测结果之前,所述跟踪目标的定位方法还包括:获取样本训练图像数据;基于深度学习框架构建卷积神经网络目标检测算法的结构;在所述卷积神经网络目标检测算法的结构中,使用所述样本训练图像数据对所述卷积神经网络目标检测算法的结构进行训练,得到所述目标预测模型。
- 如权利要求13所述的计算机设备,其特征在于,所述基于深度学习框架构建卷积神经网络目标检测算法的结构包括:获取所述卷积神经网络目标检测算法的结构的定义信息;在所述深度学习框架中按照预设的网络定义的层结构,将所述定义信息填入对应的所述卷积层和所述全连接层的层结构中,生成所述卷积神经网络目标检测算法的结构。
- 如权利要求11所述的计算机设备,其特征在于,在所述根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离之后,所述跟踪目标的定位方法还包括:将所述N个距离中小于预设阈值的所述距离的目标位置信息对应的目标区域中的待跟踪目标反馈给用户进行确认;接收所述用户返回的确认结果,并根据所述确认结果,将所述用户确认的跟踪目标作为所述原始目标点对应的跟踪目标。
- 一个或多个存储有计算机可读指令的非易失性可读存储介质,其特征在于,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行如下步骤:若接收到跟踪目标的定位请求,则获取所述定位请求中包含的用户选定的原始目标点的原始位置信息,其中,所述原始位置信息包括所述原始目标点在当前播放的视频数据中的当前帧图像和所述原始目标点在所述当前帧图像中的原始坐标;根据预设的目标预测模型,对所述当前帧图像进行目标预测,得到目标预测结果,其中,所述目标预测模型为包括两层卷积层和两层全连接层的卷积神经网络模型,所述目标预测结果包括N个目标区域和每个所述目标区域对应的目标位置信息,N为正整数,每个所述目标区域包含一个待跟踪目标;根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离;从所述N个距离中选取数值最小的距离作为目标距离,获取所述目标距离对应的目标位置信息,并将获取到的目标位置信息对应的目标区域中的待跟踪目标确定为所述原始目标点对应的跟踪目标。
- 如权利要求16所述的非易失性可读存储介质,其特征在于,所述目标区域对应的目标位置信息包括目标区域的左上角坐标、所述目标区域的宽和高,所述根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离包括:其中,i为正整数,且0<i≤N,x i为第i个所述目标位置信息对应的所述目标区域的左上角坐标的横坐标,y i为所述左上角坐标的纵坐标,w i为所述第i个所述目标位置信息对应的目标区域的宽和h i为所述第i个所述目标位置信息对应的目标区域的高;
- 如权利要求16所述的非易失性可读存储介质,其特征在于,所述根据预设的目标预测模型,对所述当前帧图像进行目标预测,得到目标预测结果之前,所述跟踪目标的定位方法还包括:获取样本训练图像数据;基于深度学习框架构建卷积神经网络目标检测算法的结构;在所述卷积神经网络目标检测算法的结构中,使用所述样本训练图像数据对所述卷积神经网络目标检测算法的结构进行训练,得到所述目标预测模型。
- 如权利要求18所述的非易失性可读存储介质,其特征在于,所述基于深度学习框架构建卷积神经网络目标检测算法的结构包括:获取所述卷积神经网络目标检测算法的结构的定义信息;在所述深度学习框架中按照预设的网络定义的层结构,将所述定义信息填入对应的所述卷积层和所述全连接层的层结构中,生成所述卷积神经网络目标检测算法的结构。
- 如权利要求16所述的非易失性可读存储介质,其特征在于,在所述根据每个所述目标区域的所述目标位置信息和所述原始坐标计算每个待跟踪目标和所述原始目标点之间的欧氏距离,得到N个距离之后,所述跟踪目标的定位方法还包括:将所述N个距离中小于预设阈值的所述距离的目标位置信息对应的目标区域中的待跟踪目标反馈给用户进行确认;接收所述用户返回的确认结果,并根据所述确认结果,将所述用户确认的跟踪目标作为所述原始目标点对应的跟踪目标。
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| SG11202101806QA SG11202101806QA (en) | 2018-11-12 | 2018-12-24 | Method, device, equipment and storage medium for locating tracked targets |
| US17/266,187 US11798174B2 (en) | 2018-11-12 | 2018-12-24 | Method, device, equipment and storage medium for locating tracked targets |
| JP2021504201A JP7060758B2 (ja) | 2018-11-12 | 2018-12-24 | 追跡ターゲットの位置決め方法、装置、機器及び記憶媒体 |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811339121.3A CN109635657B (zh) | 2018-11-12 | 2018-11-12 | 目标跟踪方法、装置、设备及存储介质 |
| CN201811339121.3 | 2018-11-12 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020098076A1 true WO2020098076A1 (zh) | 2020-05-22 |
Family
ID=66067729
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2018/123080 Ceased WO2020098076A1 (zh) | 2018-11-12 | 2018-12-24 | 跟踪目标的定位方法、装置、设备及存储介质 |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US11798174B2 (zh) |
| JP (1) | JP7060758B2 (zh) |
| CN (1) | CN109635657B (zh) |
| SG (1) | SG11202101806QA (zh) |
| WO (1) | WO2020098076A1 (zh) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111967323A (zh) * | 2020-07-16 | 2020-11-20 | 电子科技大学 | 一种基于深度学习算法的电力带电作业安全检测方法 |
| CN112489086A (zh) * | 2020-12-11 | 2021-03-12 | 北京澎思科技有限公司 | 目标跟踪方法、目标跟踪装置、电子设备及存储介质 |
| CN112866773A (zh) * | 2020-08-21 | 2021-05-28 | 海信视像科技股份有限公司 | 一种显示设备及多人场景下摄像头追踪方法 |
| CN113824880A (zh) * | 2021-08-26 | 2021-12-21 | 国网浙江省电力有限公司双创中心 | 一种基于目标检测和uwb定位的车辆跟踪方法 |
| CN113888604A (zh) * | 2021-09-27 | 2022-01-04 | 安徽清新互联信息科技有限公司 | 一种基于深度光流的目标跟踪方法 |
| CN114266806A (zh) * | 2021-11-17 | 2022-04-01 | 国网河南省电力公司电力科学研究院 | 基于目标感知深度特征压缩的电力智能巡检跟踪方法 |
| CN114339603A (zh) * | 2021-12-24 | 2022-04-12 | 珠海优特电力科技股份有限公司 | 定位区域的切换方法、装置、设备和存储介质 |
| CN114549926A (zh) * | 2022-01-24 | 2022-05-27 | 北京百度网讯科技有限公司 | 目标检测与目标检测模型的训练方法、装置 |
| CN114972421A (zh) * | 2022-04-27 | 2022-08-30 | 中南大学 | 车间物料识别追踪与定位方法、系统 |
| CN117197247A (zh) * | 2023-11-08 | 2023-12-08 | 湖南才道半导体科技有限公司 | 基于人工智能的晶圆切割道视觉定位方法 |
| CN117274310A (zh) * | 2023-09-27 | 2023-12-22 | 东风商用车有限公司 | 多目标跟踪方法、装置、设备及计算机可读存储介质 |
| CN112822450B (zh) * | 2021-01-08 | 2024-03-19 | 鹏城实验室 | 一种大规模视觉计算系统中有效节点动态遴选方法 |
| CN118230428A (zh) * | 2024-05-24 | 2024-06-21 | 浪潮软件科技有限公司 | 一种异常行为检测方法、系统及计算机可读介质 |
Families Citing this family (46)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10740620B2 (en) * | 2017-10-12 | 2020-08-11 | Google Llc | Generating a video segment of an action from a video |
| CN110110787A (zh) * | 2019-05-06 | 2019-08-09 | 腾讯科技(深圳)有限公司 | 目标的位置获取方法、装置、计算机设备及存储介质 |
| CN110363799B (zh) * | 2019-05-27 | 2021-04-06 | 浙江工业大学 | 人机共存环境下基于视觉的多运动人体目标跟踪方法 |
| CN110176027B (zh) * | 2019-05-27 | 2023-03-14 | 腾讯科技(深圳)有限公司 | 视频目标跟踪方法、装置、设备及存储介质 |
| CN112052856A (zh) * | 2019-06-05 | 2020-12-08 | 邓毅华 | 一种基于图像特征编码的目标检测方法 |
| CN110378515A (zh) * | 2019-06-14 | 2019-10-25 | 平安科技(深圳)有限公司 | 一种突发事件的预测方法、装置、存储介质和服务器 |
| CN112215762A (zh) * | 2019-07-12 | 2021-01-12 | 阿里巴巴集团控股有限公司 | 视频图像处理方法、装置及电子设备 |
| CN110490910A (zh) * | 2019-08-13 | 2019-11-22 | 顺丰科技有限公司 | 目标检测方法、装置、电子设备及存储介质 |
| CN111798487B (zh) * | 2019-08-27 | 2024-07-16 | 北京京东尚科信息技术有限公司 | 目标跟踪方法、装置和计算机可读存储介质 |
| CN110517292A (zh) * | 2019-08-29 | 2019-11-29 | 京东方科技集团股份有限公司 | 目标跟踪方法、装置、系统和计算机可读存储介质 |
| CN110634153A (zh) * | 2019-09-19 | 2019-12-31 | 上海眼控科技股份有限公司 | 目标跟踪模板更新方法、装置、计算机设备和存储介质 |
| CN110796012B (zh) * | 2019-09-29 | 2022-12-27 | 北京达佳互联信息技术有限公司 | 图像处理方法、装置、电子设备及可读存储介质 |
| CN112154444B (zh) * | 2019-10-17 | 2021-12-17 | 深圳市大疆创新科技有限公司 | 目标检测与跟踪方法、系统、可移动平台、相机及介质 |
| CN112686925A (zh) * | 2019-10-18 | 2021-04-20 | 西安光启未来技术研究院 | 一种目标追踪方法和装置 |
| CN111079525B (zh) * | 2019-11-05 | 2023-05-30 | 阿里巴巴集团控股有限公司 | 图像处理方法、设备、系统及存储介质 |
| KR20210067442A (ko) * | 2019-11-29 | 2021-06-08 | 엘지전자 주식회사 | 객체 인식을 위한 자동 레이블링 장치 및 방법 |
| CN111145214A (zh) * | 2019-12-17 | 2020-05-12 | 深圳云天励飞技术有限公司 | 目标跟踪方法、装置、终端设备及介质 |
| CN113066101B (zh) * | 2019-12-30 | 2024-11-29 | 阿里巴巴集团控股有限公司 | 数据处理方法及装置、图像处理方法及装置 |
| CN111210457B (zh) * | 2020-01-08 | 2021-04-13 | 北京天睿空间科技股份有限公司 | 一种视频分析与定位信息结合的航空器挂牌方法 |
| CN111292352B (zh) * | 2020-01-20 | 2023-08-25 | 杭州电子科技大学 | 多目标跟踪方法、装置、设备及存储介质 |
| CN113450387B (zh) * | 2020-03-26 | 2024-06-28 | 京东方科技集团股份有限公司 | 目标跟踪方法、装置、电子设备及计算机可读存储介质 |
| CN115516899A (zh) * | 2020-05-05 | 2022-12-23 | 瑞典爱立信有限公司 | 对用于无线设备寻呼的跟踪区域列表的优化 |
| CN111640140B (zh) * | 2020-05-22 | 2022-11-25 | 北京百度网讯科技有限公司 | 目标跟踪方法、装置、电子设备及计算机可读存储介质 |
| CN112257595A (zh) * | 2020-10-22 | 2021-01-22 | 广州市百果园网络科技有限公司 | 视频匹配方法、装置、设备及存储介质 |
| CN112857746A (zh) * | 2020-12-29 | 2021-05-28 | 上海眼控科技股份有限公司 | 一种灯光检测仪的追踪方法、装置、电子设备及存储介质 |
| CN112887793B (zh) * | 2021-01-25 | 2023-06-13 | 脸萌有限公司 | 视频处理方法、显示设备和存储介质 |
| CN113159032B (zh) * | 2021-04-22 | 2023-06-30 | 平安科技(深圳)有限公司 | 基于目标检测网络的目标跟踪方法、装置、设备及介质 |
| CN114089675B (zh) * | 2021-11-23 | 2023-06-09 | 长春工业大学 | 一种基于人机距离的机器控制方法及控制系统 |
| CN114310940B (zh) * | 2021-12-20 | 2024-02-06 | 三一建筑机器人(西安)研究院有限公司 | 工件定位方法、装置、介质及作业机械 |
| CN114358182A (zh) * | 2021-12-31 | 2022-04-15 | 浙江大华技术股份有限公司 | 跟踪对象的确定方法及装置、存储介质、电子装置 |
| CN114581796B (zh) * | 2022-01-19 | 2024-04-02 | 上海土蜂科技有限公司 | 目标物跟踪系统、方法及其计算机装置 |
| CN114510044A (zh) * | 2022-01-25 | 2022-05-17 | 北京圣威特科技有限公司 | Agv导航船导航方法、装置、电子设备及存储介质 |
| CN114612986B (zh) * | 2022-03-17 | 2025-10-28 | 北京市商汤科技开发有限公司 | 检测方法、装置、电子设备及存储介质 |
| CN114815877A (zh) * | 2022-03-28 | 2022-07-29 | 上海赫千电子科技有限公司 | 一种基于无人机母车的无人机目标跟踪方法 |
| CN115937247B (zh) * | 2022-08-12 | 2024-02-06 | 北京小米移动软件有限公司 | 对象跟踪的方法、装置和存储介质 |
| CN115221981B (zh) * | 2022-09-20 | 2022-12-02 | 毫末智行科技有限公司 | 一种目标跟踪方法、装置、终端设备及存储介质 |
| CN116168799A (zh) * | 2022-09-30 | 2023-05-26 | 海信视像科技股份有限公司 | 一种运动评分方法、装置和电子设备 |
| CN115830063A (zh) * | 2022-10-21 | 2023-03-21 | 中国运载火箭技术研究院 | 一种低功耗无人机自主目标跟踪方法 |
| CN115546729B (zh) * | 2022-10-24 | 2026-01-16 | 深圳市万物云科技有限公司 | 基于深度学习的遛狗违规行为检测方法、装置及设备 |
| CN115719511B (zh) * | 2022-11-08 | 2026-04-03 | 中科海微(北京)科技有限公司 | 立定跳远距离的确定方法、装置 |
| CN116434156A (zh) * | 2023-03-30 | 2023-07-14 | 大陆软件系统开发中心(重庆)有限公司 | 目标检测方法、存储介质、路侧设备及自动驾驶系统 |
| CN117036773B (zh) * | 2023-06-28 | 2025-10-28 | 华南农业大学 | 一种基于深度学习的猪只跛行检测方法和系统 |
| CN116933648B (zh) * | 2023-07-26 | 2024-12-31 | 北京航天晨信科技有限责任公司 | 一种无动力空投目标点优化方法、装置、电子设备及介质 |
| CN117274300A (zh) * | 2023-08-14 | 2023-12-22 | 四川中科友成科技有限公司 | 一种基于kcf目标跟踪算法的优化方法、装置及设备 |
| CN117854011B (zh) * | 2024-03-07 | 2024-05-03 | 福建南亿智能科技有限公司 | 一种智能ai摄像头识别对比方法及系统 |
| CN118838132B (zh) * | 2024-09-20 | 2025-03-18 | 青岛天仁微纳科技有限责任公司 | 一种纳米压印过程中模具与基底对准监测方法 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679455A (zh) * | 2017-08-29 | 2018-02-09 | 平安科技(深圳)有限公司 | 目标跟踪装置、方法及计算机可读存储介质 |
| CN108230353A (zh) * | 2017-03-03 | 2018-06-29 | 北京市商汤科技开发有限公司 | 目标跟踪方法、系统及电子设备 |
| CN108268823A (zh) * | 2016-12-30 | 2018-07-10 | 纳恩博(北京)科技有限公司 | 目标再识别方法和装置 |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5919665B2 (ja) * | 2011-07-19 | 2016-05-18 | 日本電気株式会社 | 情報処理装置、物体追跡方法および情報処理プログラム |
| JP6614611B2 (ja) | 2016-02-29 | 2019-12-04 | Kddi株式会社 | 画像間類似度を考慮して物体を追跡する装置、プログラム及び方法 |
| CN106803263A (zh) * | 2016-11-29 | 2017-06-06 | 深圳云天励飞技术有限公司 | 一种目标跟踪方法及装置 |
| EP3545466A4 (en) * | 2016-12-29 | 2019-11-27 | Zhejiang Dahua Technology Co., Ltd. | SYSTEMS AND METHOD FOR DETECTING OBJECTS IN PICTURES |
| CN107292911B (zh) * | 2017-05-23 | 2021-03-30 | 南京邮电大学 | 一种基于多模型融合和数据关联的多目标跟踪方法 |
| US10679428B1 (en) * | 2017-05-26 | 2020-06-09 | Snap Inc. | Neural network-based image stream modification |
| US20190130189A1 (en) * | 2017-10-30 | 2019-05-02 | Qualcomm Incorporated | Suppressing duplicated bounding boxes from object detection in a video analytics system |
| CN108363997A (zh) * | 2018-03-20 | 2018-08-03 | 南京云思创智信息科技有限公司 | 一种在视频中对特定人的实时跟踪方法 |
| CN108765455B (zh) * | 2018-05-24 | 2021-09-21 | 中国科学院光电技术研究所 | 一种基于tld算法的目标稳定跟踪方法 |
-
2018
- 2018-11-12 CN CN201811339121.3A patent/CN109635657B/zh active Active
- 2018-12-24 US US17/266,187 patent/US11798174B2/en active Active
- 2018-12-24 WO PCT/CN2018/123080 patent/WO2020098076A1/zh not_active Ceased
- 2018-12-24 SG SG11202101806QA patent/SG11202101806QA/en unknown
- 2018-12-24 JP JP2021504201A patent/JP7060758B2/ja active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108268823A (zh) * | 2016-12-30 | 2018-07-10 | 纳恩博(北京)科技有限公司 | 目标再识别方法和装置 |
| CN108230353A (zh) * | 2017-03-03 | 2018-06-29 | 北京市商汤科技开发有限公司 | 目标跟踪方法、系统及电子设备 |
| CN107679455A (zh) * | 2017-08-29 | 2018-02-09 | 平安科技(深圳)有限公司 | 目标跟踪装置、方法及计算机可读存储介质 |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111967323B (zh) * | 2020-07-16 | 2022-07-26 | 电子科技大学 | 一种基于深度学习算法的电力带电作业安全检测方法 |
| CN111967323A (zh) * | 2020-07-16 | 2020-11-20 | 电子科技大学 | 一种基于深度学习算法的电力带电作业安全检测方法 |
| CN112866773A (zh) * | 2020-08-21 | 2021-05-28 | 海信视像科技股份有限公司 | 一种显示设备及多人场景下摄像头追踪方法 |
| CN112866773B (zh) * | 2020-08-21 | 2023-09-26 | 海信视像科技股份有限公司 | 一种显示设备及多人场景下摄像头追踪方法 |
| CN112489086A (zh) * | 2020-12-11 | 2021-03-12 | 北京澎思科技有限公司 | 目标跟踪方法、目标跟踪装置、电子设备及存储介质 |
| CN112822450B (zh) * | 2021-01-08 | 2024-03-19 | 鹏城实验室 | 一种大规模视觉计算系统中有效节点动态遴选方法 |
| CN113824880A (zh) * | 2021-08-26 | 2021-12-21 | 国网浙江省电力有限公司双创中心 | 一种基于目标检测和uwb定位的车辆跟踪方法 |
| CN113888604A (zh) * | 2021-09-27 | 2022-01-04 | 安徽清新互联信息科技有限公司 | 一种基于深度光流的目标跟踪方法 |
| CN114266806A (zh) * | 2021-11-17 | 2022-04-01 | 国网河南省电力公司电力科学研究院 | 基于目标感知深度特征压缩的电力智能巡检跟踪方法 |
| CN114266806B (zh) * | 2021-11-17 | 2025-10-17 | 国网河南省电力公司电力科学研究院 | 基于目标感知深度特征压缩的电力智能巡检跟踪方法 |
| CN114339603B (zh) * | 2021-12-24 | 2023-08-25 | 珠海优特电力科技股份有限公司 | 定位区域的切换方法、装置、设备和存储介质 |
| CN114339603A (zh) * | 2021-12-24 | 2022-04-12 | 珠海优特电力科技股份有限公司 | 定位区域的切换方法、装置、设备和存储介质 |
| CN114549926A (zh) * | 2022-01-24 | 2022-05-27 | 北京百度网讯科技有限公司 | 目标检测与目标检测模型的训练方法、装置 |
| CN114972421A (zh) * | 2022-04-27 | 2022-08-30 | 中南大学 | 车间物料识别追踪与定位方法、系统 |
| CN117274310A (zh) * | 2023-09-27 | 2023-12-22 | 东风商用车有限公司 | 多目标跟踪方法、装置、设备及计算机可读存储介质 |
| CN117197247A (zh) * | 2023-11-08 | 2023-12-08 | 湖南才道半导体科技有限公司 | 基于人工智能的晶圆切割道视觉定位方法 |
| CN117197247B (zh) * | 2023-11-08 | 2024-03-29 | 湖南才道半导体科技有限公司 | 基于人工智能的晶圆切割道视觉定位方法 |
| CN118230428A (zh) * | 2024-05-24 | 2024-06-21 | 浪潮软件科技有限公司 | 一种异常行为检测方法、系统及计算机可读介质 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109635657A (zh) | 2019-04-16 |
| JP7060758B2 (ja) | 2022-04-26 |
| CN109635657B (zh) | 2023-01-06 |
| JP2021531592A (ja) | 2021-11-18 |
| US20210295536A1 (en) | 2021-09-23 |
| US11798174B2 (en) | 2023-10-24 |
| SG11202101806QA (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020098076A1 (zh) | 跟踪目标的定位方法、装置、设备及存储介质 | |
| CN112150551B (zh) | 物体位姿的获取方法、装置和电子设备 | |
| WO2022170742A1 (zh) | 目标检测方法、装置、电子设备和存储介质 | |
| CN107430686B (zh) | 用于移动设备定位的区域描述文件的众包创建和更新 | |
| CN109448018B (zh) | 跟踪目标的定位方法、装置、设备及存储介质 | |
| WO2020143323A1 (zh) | 遥感影像分割方法、装置及存储介质、服务器 | |
| CN112528978B (zh) | 人脸关键点的检测方法、装置、电子设备及存储介质 | |
| US12073507B2 (en) | Point-based neural radiance field for three dimensional scene representation | |
| WO2020103647A1 (zh) | 物体关键点的定位方法、图像处理方法、装置及存储介质 | |
| WO2022089391A1 (zh) | 模型训练方法、人体姿态检测方法、装置、设备及存储介质 | |
| CN109858333B (zh) | 图像处理方法、装置、电子设备及计算机可读介质 | |
| WO2021249114A1 (zh) | 目标跟踪方法和目标跟踪装置 | |
| CN111160288A (zh) | 手势关键点检测方法、装置、计算机设备和存储介质 | |
| JP7488846B2 (ja) | 連合学習機構を利用した画像IoTプラットフォーム | |
| CN115359092B (zh) | 一种注视点预测模型的训练方法、装置及电子设备 | |
| WO2021103945A1 (zh) | 地图融合方法及装置、设备、存储介质 | |
| CN113284237A (zh) | 一种三维重建方法、系统、电子设备及存储介质 | |
| JP2022020588A (ja) | アクティブインタラクションの方法、装置、電子デバイス及び可読記憶媒体 | |
| CN112733641A (zh) | 物体尺寸测量方法、装置、设备及存储介质 | |
| CN114565916A (zh) | 目标检测模型训练方法、目标检测方法以及电子设备 | |
| CN120014209B (zh) | 智能体三维虚拟场景的生成优化方法、装置、设备及介质 | |
| WO2022063321A1 (zh) | 图像处理方法、装置、设备及存储介质 | |
| CN110517221A (zh) | 基于现实坐标的缝隙定位方法、装置和存储介质 | |
| CN115063473A (zh) | 物体高度的检测方法、装置、计算机设备、存储介质 | |
| US20240029427A1 (en) | Method for determining unmanned aerial vehicle acquisition viewpoints, computer apparatus, and storage medium |
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: 18940118 Country of ref document: EP Kind code of ref document: A1 |
|
| ENP | Entry into the national phase |
Ref document number: 2021504201 Country of ref document: JP Kind code of ref document: A |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 20.08.2021) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 18940118 Country of ref document: EP Kind code of ref document: A1 |