Target recognition method and appartus, storage medium, and electronic device

Abstract

A method for identifying a target, a non-transitory computer-readable storage medium, and an electronic device include: acquiring a first image and a second image, the first image and the second image each including a target to be determined; generating a prediction path based on the first image and the second image, both ends of the prediction path respectively corresponding to the first image and the second image; and performing validity determination on the prediction path and determining, according to a determination result, whether the targets to be determined in the first image and the second image are the same target to be determined.

What is claimed is: 
     
         1 . A method for identifying a target, comprising:
 acquiring a first image and a second image, the first image and the second image each comprising a target to be determined;   generating a prediction path based on the first image and the second image, both ends of the prediction path respectively corresponding to targets to be determined in the first image and the second image; and   performing validity determination on the prediction path, and determining, according to a determination result, whether the targets to be determined in the first image and the second image are the same target to be determined,   wherein performing the validity determination on the prediction path and determining, according to the determination result, whether the targets to be determined in the first image and the second image are the same target to be determined comprises:   performing, through a neural network, validity determination on the prediction path and determining, according to the determination result, whether the targets to be determined in the first image and the second image are the same target to be determined;   wherein performing, through the neural network, the validity determination on the prediction path and determining whether the targets to be determined in the first image and the second image are the same target to be determined according to the determination result comprises:   acquiring a temporal difference between adjacent images in the prediction path according to temporal information of the adjacent images; acquiring a spatial difference between the adjacent images according to spatial information of the adjacent images; and acquiring a feature difference between the targets to be determined in the adjacent images according to feature information of the targets to be determined in the adjacent images;   inputting the obtained temporal difference, spatial difference, and feature difference between the adjacent images in the prediction path into a Long Short-Term Memory (LSTM) network to obtain an identification probability of the targets to be determined in the prediction path; and   determining, according to the identification probability of the targets to be determined in the prediction path, whether the targets to be determined in the first image and the second image are the same target to be determined.   
     
     
         2 . The method according to  claim 1 , wherein before generating the prediction path based on the first image and the second image, the method further comprises:
 determining a preliminary sameness probability value of the targets to be determined respectively contained in the first image and the second image according to temporal information, spatial information, and image feature information of the first image and temporal information, spatial information, and image feature information of the second image;   wherein generating the prediction path based on the first image and the second image comprises:   generating the prediction path based on the first image and the second image if the preliminary sameness probability value is greater than a preset value.   
     
     
         3 . The method according to  claim 2 , wherein determining the preliminary sameness probability value of the targets to be determined respectively contained in the first image and the second image according to the temporal information, the spatial information, and the image feature information of the first image and the temporal information, the spatial information, and the image feature information of the second image comprises:
 inputting the first image, the second image, and a difference in temporal information and a difference in spatial information between the first image and the second image into a Siamese Convolutional Neural Network (Siamese-CNN) to obtain the preliminary sameness probability value of the targets to be determined in the first image and the second image.   
     
     
         4 . The method according to  claim 1 , wherein generating the prediction path based on the first image and the second image comprises:
 generating, through a probability model, the prediction path of the targets to be determined according to the feature information of the first image, the temporal information of the first image, the spatial information of the first image, the feature information of the second image, the temporal information of the second image, and the spatial information of the second image.   
     
     
         5 . The method according to  claim 4 , wherein generating, through the probability model, the prediction path of the targets to be determined comprises:
 determining, through a chain Markov Random Field (MRF) model, all images comprising information of the targets to be determined and having a spatiotemporal sequence relationship with the first image and the second image from an acquired image set; and   generating, according to temporal information and spatial information corresponding to all the determined images, the prediction path of the targets to be determined;   wherein generating, according to the temporal information and the spatial information corresponding to all the determined images, the prediction path of the targets to be determined comprises:   generating a prediction path with the first image as a head node and the second image as a tail node according to the temporal information and the spatial information corresponding to all the determined images, wherein the prediction path further corresponds to at least one intermediate node in addition to the head node and the tail node.   
     
     
         6 . The method according to  claim 5 , wherein determining, through the chain MRF model, all images comprising the information of the targets to be determined and having the spatiotemporal sequence relationship with the first image and the second image from the acquired image set comprises:
 acquiring position information of all camera devices from a start position to an end position by using a position corresponding to the spatial information of the first image as the start position and using a position corresponding to the spatial information of the second image as the end position;   generating, according to the relationships between positions indicated by the position information of all the camera devices, at least one device path by using a camera device corresponding to the start position as a start point and using a camera device corresponding to the end position as an end point, wherein each device path further comprises information of at least one other camera device in addition to the camera device as the start point and the camera device as the end point; and   determining, from images captured by each of the other camera devices on the current path, an image for each device path by using time corresponding to the temporal information of the first image as start time and using time corresponding to the temporal information of the second image as end time, wherein the image comprises the information of the targets to be determined, and has a set temporal sequence relationship with an image which comprises the information of the targets to be determined and is captured by a previous camera device adjacent to the current camera device.   
     
     
         7 . The method according to  claim 6 , wherein generating the prediction path with the first image as the head node and the second image as the tail node according to the temporal information and the spatial information corresponding to all the determined images comprises:
 generating, according to the temporal sequence relationship of the determined images, a plurality of connected intermediate nodes having a spatiotemporal sequence relationship for each device path, and generating, according to the head node, the tail node, and the intermediate nodes, an image path having a spatiotemporal sequence relationship and corresponding to the current device path; and   determining, from the image path corresponding to each device path, a maximum probability image path with the first image as the head node and the second image as the tail node as the prediction path of the targets to be determined.   
     
     
         8 . The method according to  claim 7 , wherein determining, from the image path corresponding to each device path, the maximum probability image path with the first image as the head node and the second image as the tail node as the prediction path of the targets to be determined comprises:
 acquiring, for the image path corresponding to each device path, a probability of images of every two adjacent nodes in the image path having information of the same target to be determined;   calculating, according to the probability of the images of every two adjacent nodes in the image path having the information of the same target to be determined, a probability of the image path being a prediction path of the target to be determined; and   determining, according to the probability of each image path being a prediction path of the target to be determined, the maximum probability image path as the prediction path of the target to be determined.   
     
     
         9 . The method according to  claim 1 , wherein acquiring the feature difference between the targets to be determined in the adjacent images according to the feature information of the targets to be determined in the adjacent images comprises:
 separately acquiring feature information of the targets to be determined in the adjacent images through the Siamese-CNN; and   acquiring the feature difference between the targets to be determined in the adjacent images according to the separately acquired feature information.   
     
     
         10 . An apparatus for identifying a target, comprising:
 a processor; and   a memory for storing instructions executable by the processor;   wherein the processor is configured to:   acquire a first image and a second image, the first image and the second image each comprising a target to be determined;   generate a prediction path based on the first image and the second image, both ends of the prediction path respectively corresponding to targets to be determined in the first image and the second image; and   perform validity determination on the prediction path and determine,   according to a determination result, whether the targets to be determined in the first image and the second image are the same target to be determined,   wherein the processor is specifically configured to:   perform validity determination on the prediction path through a neural network and determine, according to a determination result, whether the targets to be determined in the first image and the second image are the same target to be determined;   wherein the operation of performing the validity determination on the prediction path through the neural network and determine, according to the determination result, whether the targets to be determined in the first image and the second image are the same target to be determined comprises:   acquiring a temporal difference between adjacent images in the prediction path according to temporal information of the adjacent images; acquire a spatial difference between the adjacent images according to spatial information of the adjacent images;   and acquire a feature difference between the targets to be determined in the adjacent images according to feature information of the targets to be determined in the adjacent images;   inputting the obtained temporal difference, spatial difference, and feature difference between the adjacent images in the prediction path into a Long Short-Term Memory (LSTM) network to obtain an identification probability of the targets to be determined in the prediction path; and   determining, according to the identification probability of the targets to be determined in the prediction path, whether the targets to be determined in the first image and the second image are the same target to be determined.   
     
     
         11 . The apparatus according to  claim 10 , wherein the processor is configured to:
 determine, according to temporal information, spatial information, and image feature information of the first image and temporal information, spatial information, and image feature information of the second image, a preliminary sameness probability value of the targets to be determined respectively contained in the first image and the second image;   wherein the operation of generating the prediction path based on the first image and the second image comprises:   generating the prediction path based on the first image and the second image if the preliminary sameness probability value is greater than a preset value.   
     
     
         12 . The apparatus according to  claim 11 , wherein the processor is configured to:
 input the first image, the second image, and a difference in temporal information and a difference in spatial information between the first image and the second image into a Siamese Convolutional Neural Network (Siamese-CNN) to obtain a preliminary sameness probability value of the targets to be determined in the first image and the second image.   
     
     
         13 . The apparatus according to  claim 10 , wherein the processor is configured to:
 generate the prediction path of the targets to be determined through a probability model according to the feature information of the first image, the temporal information of the first image, the spatial information of the first image, the feature information of the second image, the temporal information of the second image, and the spatial information of the second image.   
     
     
         14 . The apparatus according to  claim 13 , wherein the processor is configured to:
 determine, through a chain Markov Random Field (MRF) model, all images comprising information of the targets to be determined and having a spatiotemporal sequence relationship with the first image and the second image from an acquired image set; and   generate, according to temporal information and spatial information corresponding to all the determined images, the prediction path of the targets to be determined;   wherein the operation of generating, according to the temporal information and the spatial information corresponding to all the determined images, the prediction path of the targets to be determined comprises:   generating a prediction path with the first image as a head node and the second image as a tail node according to the temporal information and the spatial information corresponding to all the determined images, wherein the prediction path further corresponds to at least one intermediate node in addition to the head node and the tail node.   
     
     
         15 . The apparatus according to  claim 14 , wherein the processor is further configured to:
 acquire position information of all camera devices from a start position to an end position by using a position corresponding to the spatial information of the first image as the start position and using a position corresponding to the spatial information of the second image as the end position;   generate, according to the relationships between positions indicated by the position information of all the camera devices, at least one device path by using a camera device corresponding to the start position as a start point and using a camera device corresponding to the end position as an end point, wherein each device path further comprises information of at least one other camera device in addition to the camera device as the start point and the camera device as the end point; and   determine, from images captured by each of the other camera devices on the current path, an image for each device path by using time corresponding to the temporal information of the first image as start time and using time corresponding to the temporal information of the second image as end time, wherein the image comprises the information of the targets to be determined, and has a set temporal sequence relationship with an image which comprises the information of the targets to be determined and is captured by a previous camera device adjacent to the current camera device.   
     
     
         16 . The apparatus according to  claim 15 , wherein the processor is configured to:
 generate, according to the temporal sequence relationship of the determined images, a plurality of connected intermediate nodes having a spatiotemporal sequence relationship for each device path; and generate, according to the head node, the tail node, and the intermediate nodes, an image path having a spatiotemporal sequence relationship and corresponding to the current device path; and   determine, from the image path corresponding to each device path, a maximum probability image path with the first image as the head node and the second image as the tail node as the prediction path of the targets to be determined.   
     
     
         17 . The apparatus according to  claim 16 , wherein the processor is further configured to:
 acquire, for the image path corresponding to each device path, a probability of images of every two adjacent nodes in the image path having information of the same target to be determined;   calculate, according to the probability of the images of every two adjacent nodes in the image path having the information of the same target to be determined, a probability of the image path being a prediction path of the target to be determined; and   determine, according to the probability of each image path being a prediction path of the target to be determined, the maximum probability image path as the prediction path of the target to be determined.   
     
     
         18 . The apparatus according to  claim 10 , wherein the operation of acquiring the feature difference between the targets to be determined in the adjacent images according to the feature information of the targets to be determined in the adjacent images comprises:
 separately acquiring feature information of the targets to be determined in the adjacent images through the Siamese-CNN; and   acquiring the feature difference between the targets to be determined in the adjacent images according to the separately acquired feature information.   
     
     
         19 . A non-transitory computer-readable storage medium, having computer program instructions stored thereon, wherein the program instructions, when being executed by a processor, are configured to perform the operations of:
 acquiring a first image and a second image, the first image and the second image each comprising a target to be determined;   generating a prediction path based on the first image and the second image, both ends of the prediction path respectively corresponding to targets to be determined in the first image and the second image; and   performing validity determination on the prediction path, and determining, according to a determination result, whether the targets to be determined in the first image and the second image are the same target to be determined,   wherein the operation of performing the validity determination on the prediction path and determining, according to the determination result, whether the targets to be determined in the first image and the second image are the same target to be determined comprises:   performing, through a neural network, validity determination on the prediction path and determining, according to the determination result, whether the targets to be determined in the first image and the second image are the same target to be determined; and   wherein performing, through the neural network, the validity determination on the prediction path and determining whether the targets to be determined in the first image and the second image are the same target to be determined according to the determination result comprises:   acquiring a temporal difference between adjacent images in the prediction path according to temporal information of the adjacent images; acquiring a spatial difference between the adjacent images according to spatial information of the adjacent images; and acquiring a feature difference between the targets to be determined in the adjacent images according to feature information of the targets to be determined in the adjacent images;   inputting the obtained temporal difference, spatial difference, and feature difference between the adjacent images in the prediction path into a Long Short-Term Memory (LSTM) network to obtain an identification probability of the targets to be determined in the prediction path; and   determining, according to the identification probability of the targets to be determined in the prediction path, whether the targets to be determined in the first image and the second image are the same target to be determined.   
     
     
         20 . The non-transitory computer-readable storage medium according to  claim 19 , wherein before generating the prediction path based on the first image and the second image, the program instructions, when being executed by the processor, are further configured to perform the operation of:
 determining a preliminary sameness probability value of the targets to be determined respectively contained in the first image and the second image according to temporal information, spatial information, and image feature information of the first image and temporal information, spatial information, and image feature information of the second image;   wherein generating the prediction path based on the first image and the second image comprises:   generating the prediction path based on the first image and the second image if the preliminary sameness probability value is greater than a preset value.