Target detection method and system based on data fusion

Abstract

The invention discloses a target detection method and system based on data fusion, and the method comprises the steps: carrying out the target detection of a historical point cloud data set through employing all agents, and constructing a graph matching matrix based on the similarity features among all obtained target detection results; obtaining each initial vertex feature based on the graph matching matrix; processing an edge weight matrix of the graph matching matrix, a degree matrix of the graph matching matrix and the initial vertex feature by adopting a first graph convolutional network to obtain an updated vertex feature; performing matching analysis on each updated vertex feature and the other updated vertex features by adopting a second graph convolutional network to obtain a double random matrix; constructing a loss function, and training the graph convolutional neural network model by using the loss function; and processing a target detection result of the target agent by using the trained model, and fusing the target detection result. According to the method provided by the embodiment of the invention, the accuracy of target detection is improved.

Description

Technical Field

[0001] The present invention relates to the field of detection technology, and in particular to a target detection method and system based on data fusion. Background Art

[0002] As the cornerstone of computer vision, target detection aims to identify specific targets in images or videos and determine the location of the targets. It is an important technology in the field of computer vision. Target detection gives machines the ability to recognize and understand image content, and plays a vital role in security monitoring, autonomous driving, smart retail and other fields.

[0003] However, since the captured images or videos may be interfered by various factors, the image quality is low, and in the captured images, the target to be detected may also be blocked by other objects, resulting in low accuracy of target detection.

[0004] It can be seen that how to improve the accuracy of target detection has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the invention

[0005] The present invention provides a target detection method and system based on data fusion to solve the technical problem of low accuracy of target detection and to achieve the effect of improving the accuracy of target detection.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a target detection method based on data fusion, including:

[0007] All agents are used to perform target detection on the historical point cloud dataset, and a graph matching matrix is ​​constructed based on the similarity features between all target detection results.

[0008] Acquire each initial vertex feature based on the graph matching matrix; use a first graph convolutional network to process an edge weight matrix of the graph matching matrix, a degree matrix of the graph matching matrix, and the initial vertex features to obtain updated vertex features;

[0009] Using a second graph convolutional network to perform matching analysis on each of the updated vertex features and the remaining updated vertex features to obtain a non-negative score matrix; normalizing the non-negative score matrix to obtain a double random matrix; constructing a loss function based on the difference characteristics between the double random matrix and the true matching matrix, and training the graph convolutional neural network model with the loss function;

[0010] The trained graph convolutional neural network model is used to process all real-time target detection results of the target intelligent agent to obtain each associated target detection result of the target intelligent agent, and all the associated target detection results are weightedly fused to obtain a fused target detection result.

[0011] As one of the preferred solutions, the step of constructing a graph matching matrix based on similarity features among all target detection results obtained includes:

[0012] Taking each of the target detection results as a node, connecting each node to the remaining nodes, and taking the line between any two nodes as an edge;

[0013] Constructing a first feature vector of each node based on the label of each target detection result, the shape of the detection box, and the position of the center point of the detection box, and constructing a second feature vector of each edge based on the length and angle of each edge and the label of each corresponding node;

[0014] Calculating the comprehensive node similarity between the corresponding nodes according to the similarity feature between the first feature vectors of two nodes, and calculating the comprehensive edge line similarity between the corresponding edges according to the similarity feature between the second feature vectors of two edges;

[0015] The comprehensive node similarity is used as an element at a diagonal position of the graph matching matrix, and the comprehensive edge similarity is used as an element at a non-diagonal position of the graph matching matrix.

[0016] As one preferred solution, the step of calculating the comprehensive node similarity between corresponding nodes according to the similarity feature between the first feature vectors of two nodes includes:

[0017] The similarity of the shapes of the detection boxes of the two nodes is used as the first node similarity, the Euclidean distance between the positions of the center points of the detection boxes of the two nodes is inversely proportionally processed to obtain the second node similarity, the similarity of the labels of the two nodes is used as the third node similarity, and the node similarity of the two nodes is calculated based on the linear relationship between the first node similarity, the second node similarity and the third node similarity;

[0018] The calculating the comprehensive edge line similarity between the corresponding edges according to the similarity feature between the second feature vectors of the two edges comprises:

[0019] The similarity of the labels of the starting nodes of the two edges is taken as the first edge point similarity, and the similarity of the labels of the ending nodes of the two edges is taken as the second edge point similarity. The third edge similarity is obtained based on the difference characteristics of the lengths of the two edges; the fourth edge similarity is obtained based on the difference characteristics of the angles of the two edges. According to the linear relationship between the first edge point similarity, the second edge point similarity, the third edge similarity and the fourth edge similarity, the comprehensive edge similarity of the two nodes is calculated.

[0020] As one of the preferred solutions, the step of obtaining each initial vertex feature based on the graph matching matrix includes:

[0021] The elements at all diagonal positions of the graph matching matrix are used as initial vertex features.

[0022] As one of the preferred solutions, the first graph convolutional network includes two fully connected layers, and the activation function of the first graph convolutional network is a ReLU activation function;

[0023] The step of using the first graph convolutional network to process the edge weight matrix of the graph matching matrix, the degree matrix of the graph matching matrix, and the initial vertex features to obtain updated vertex features includes:

[0024] m (k) =D -1 Wf1(v (k-1) )+f1(v (k-1) )

[0025] Among them, m (k) Update the k-th vertex feature, D -1 is the inverse matrix of the degree matrix of the graph matching matrix, W is the edge weight matrix of the graph matching matrix, f1 is the fully connected layer of the first graph convolutional network, v (k-1) is the k-1th initial vertex feature.

[0026] As one of the preferred solutions, the second graph convolutional network includes a fully connected layer, and the activation function of the second graph convolutional network is an exponential activation function.

[0027] As one of the preferred solutions, the sum of the elements in each row of the double random matrix is ​​1, and the sum of the elements in each column of the double random matrix is ​​1.

[0028] As one of the preferred solutions, the loss function is constructed based on the difference characteristics between the double random matrix and the true matching matrix, including:

[0029]

[0030] Among them, L is the loss function, np is the number of updated vertex features, log is the logarithmic function, P i,j is the element value of the i-th row and j-th column in the doubly random matrix, X i,j is the element value of the i-th row and j-th column in the true matching matrix.

[0031] As one of the preferred solutions, weighted fusion of all the associated target detection results to obtain a fused target detection result includes:

[0032] The classification confidence of each detection frame of the target intelligent body is used as the corresponding weight, and the associated target detection results of each detection frame of the target intelligent body are weighted and summed as the fused target detection result.

[0033] Another embodiment of the present invention provides a target detection system based on data fusion, including:

[0034] A processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, a target detection method based on data fusion as described in any one of the above items is implemented.

[0035] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0036] (1) All agents are used to perform target detection on the historical point cloud dataset. A graph matching matrix is ​​constructed based on the similarity features between all target detection results. The graph matching matrix records the similarity between multiple detection results of different agents, thereby retaining the most original correlation between the detection results.

[0037] (2) Based on the graph matching matrix, each initial vertex feature is obtained. Since the initial feature only records the original node information, when there is a position error in the detection result, the node similarity will decrease, resulting in a weakened correlation between different detection results of the same target. Therefore, in order to reduce the negative impact of the error, the edge similarity information is introduced to correct the vertex features. The first graph convolutional network is used to process the edge weight matrix of the graph matching matrix, the degree matrix of the graph matching matrix and the initial vertex features to obtain updated vertex features. The features of adjacent vertices are learned through the graph convolutional network to update its own features, thereby improving the accuracy of subsequent association fusion analysis and enhancing the robustness of target detection in complex environments, thereby improving the accuracy of target detection;

[0038] (3) A second graph convolutional network is used to perform matching analysis on each updated vertex feature and the remaining updated vertex features to obtain a non-negative score matrix. The target association is converted into a vertex classification problem and solved using a graph convolutional network. The non-negative score matrix is ​​normalized to obtain a double random matrix to meet the one-to-one constraint of the graph matching problem. The elements in the double random matrix represent the matching probability between nodes. A loss function is constructed based on the difference between the double random matrix and the true matching matrix. The graph convolutional neural network model is trained using the loss function. The back-propagation algorithm is used to perform gradient descent to optimize the graph matching results. By minimizing the loss function, the model learns how to generate a prediction probability matrix that is more consistent with the true matching label, thereby gradually improving the matching performance and ensuring the robustness of target detection in complex environments.

[0039] (4) The trained graph convolutional neural network model is used to process all real-time target detection results of the target agent to obtain the detection results of each associated target of the target agent. All associated target detection results are weighted fused to obtain the fused target detection results. The results of multiple detection boxes are integrated through a weighted fusion strategy to further improve the accuracy of target detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flowchart of a target detection method based on data fusion in one embodiment of the present invention;

[0041] Figure 2 It is a schematic diagram of obtaining the loss function in one embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0044] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0045] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0046] An embodiment of the present invention provides a target detection method based on data fusion. For details, see Figure 1 , Figure 1 The flowchart of the target detection method based on data fusion in one embodiment of the present invention is shown as follows:

[0047] S1: Use all agents to perform target detection on the historical point cloud dataset, and build a graph matching matrix based on the similarity features between all target detection results;

[0048] S2: acquiring each initial vertex feature based on the graph matching matrix; using a first graph convolutional network to process an edge weight matrix of the graph matching matrix, a degree matrix of the graph matching matrix, and the initial vertex features to obtain updated vertex features;

[0049] S3: using a second graph convolutional network to perform matching analysis on each of the updated vertex features and the remaining updated vertex features to obtain a non-negative score matrix; normalizing the non-negative score matrix to obtain a double random matrix; constructing a loss function based on the difference characteristics between the double random matrix and the true matching matrix, and training the graph convolutional neural network model with the loss function;

[0050] S4: Use the trained graph convolutional neural network model to process all real-time target detection results of the target intelligent agent to obtain each associated target detection result of the target intelligent agent, perform weighted fusion on all the associated target detection results, and obtain a fused target detection result.

[0051] In step S1, a graph matching matrix is ​​constructed based on the similarity features between all target detection results obtained, specifically including: taking each target detection result as a node, connecting each node to the remaining nodes, and taking the line between any two nodes as an edge;

[0052] The first feature vector of each node is constructed based on the label of each target detection result, the shape of the detection box, and the position of the center point of the detection box. The second feature vector of each edge is constructed based on the length and angle of each edge and the label of each corresponding node.

[0053] Calculate the comprehensive node similarity between the corresponding nodes according to the similarity feature between the first eigenvectors of two nodes, and calculate the comprehensive edge similarity between the corresponding edges according to the similarity feature between the second eigenvectors of two edges;

[0054] The comprehensive node similarity is taken as the element on the diagonal position of the graph matching matrix, and the comprehensive edge similarity is taken as the element on the non-diagonal position of the graph matching matrix.

[0055] Among them, the comprehensive node similarity between the corresponding nodes is calculated according to the similarity characteristics between the first feature vectors of the two nodes, including: taking the similarity of the detection box shapes of the two nodes as the first node similarity, performing inverse proportional processing on the Euclidean distance between the positions of the center points of the detection boxes of the two nodes to obtain the second node similarity, taking the similarity of the labels of the two nodes as the third node similarity, and calculating the node similarity of the two nodes based on the linear relationship between the first node similarity, the second node similarity and the third node similarity.

[0056] The comprehensive edge line similarity between the corresponding edges is calculated according to the similarity features between the second eigenvectors of the two edges, including: taking the similarity of the labels of the starting nodes of the two edges as the first edge line point similarity, taking the similarity of the labels of the ending nodes of the two edges as the second edge line point similarity, obtaining the third edge line similarity based on the difference features of the lengths of the two edges; obtaining the fourth edge line similarity based on the difference features of the angles of the two edges, and calculating the comprehensive edge line similarity of the two nodes according to the linear relationship between the first edge line point similarity, the second edge line point similarity, the third edge line similarity and the fourth edge line similarity.

[0057] Specifically, each node represents a target detected by the agent, and the node features include the target's location coordinates, detection box shape, detection box deflection angle, predicted category, and predicted probability distribution. All nodes are connected in a fully connected manner to construct edges between nodes. The features of the edges include the relative distance and angle between nodes, as well as the label information of the start and end points.

[0058] Detection results with the same label, similar detection box shapes and close distances will have higher node similarity. Based on this, the node similarity and edge similarity are calculated, and the calculated node and edge similarity information is integrated into the graph matching matrix.

[0059] The elements on all diagonal positions of the graph matching matrix are used as initial vertex features, and the first graph convolutional network is used to process the edge weight matrix of the graph matching matrix, the degree matrix of the graph matching matrix and the initial vertex features to obtain updated vertex features.

[0060] Specifically, the first graph convolutional network includes two fully connected layers, and the activation function of the first graph convolutional network is the ReLU activation function.

[0061] The first graph convolutional network is used to process the edge weight matrix of the graph matching matrix, the degree matrix of the graph matching matrix and the initial vertex features to obtain updated vertex features, including:

[0062] m (k) =D -1 Wf1(v (k-1) )+f1(v (k-1) )

[0063] Among them, m (k) Update the k-th vertex feature, D -1 is the inverse matrix of the degree matrix of the graph matching matrix, W is the edge weight matrix of the graph matching matrix, f1 is the fully connected layer of the first graph convolutional network, v (k-1) is the k-1th initial vertex feature.

[0064] The second graph convolutional network is used to perform matching analysis on each updated vertex feature and the remaining updated vertex features to obtain a non-negative score matrix; the non-negative score matrix is ​​normalized to obtain a double random matrix; a loss function is constructed based on the difference characteristics between the double random matrix and the true matching matrix, and the graph convolutional neural network model is trained with the loss function.

[0065] The second graph convolutional network includes a fully connected layer, and the activation function of the second graph convolutional network is an exponential activation function.

[0066] The non-negative score matrix usually represents some form of similarity or association, where each element is non-negative. In order to convert it into a double random matrix, we need to perform the following two steps of normalization:

[0067] Each row of the matrix is ​​normalized so that the sum of the elements in each row is 1. The row-normalized matrix represents the relative relevance of each node to other nodes.

[0068] On the basis of row normalization, each column of the matrix is ​​normalized so that the sum of the elements in each column is also 1. Column normalization ensures that the distribution of the association degree starting from each node is also uniform.

[0069] After these two steps of normalization, the original non-negative score matrix is ​​transformed into a double random matrix.

[0070] The difference between the double random matrix and the true matching matrix is ​​the basis for constructing the loss function. The true matching matrix is ​​usually a binary matrix that represents the true matching relationship between nodes (such as 1 for matching and 0 for mismatching).

[0071] Specifically include:

[0072]

[0073] Among them, L is the loss function, n p is the number of updated vertex features, log is the logarithmic function, P i,j is the element value of the i-th row and j-th column in the doubly random matrix, X i,j is the element value of the i-th row and j-th column in the true matching matrix.

[0074] In step S4, all real-time target detection results of the target intelligent agent are processed using the trained graph convolutional neural network model to obtain the associated target detection results of the target intelligent agent, and all associated target detection results are weighted fused to obtain the fused target detection result. The specific steps include: taking the classification confidence of each detection frame of the target intelligent agent as the corresponding weight, and performing weighted summation of the associated target detection results of each detection frame of the target intelligent agent as the fused target detection result.

[0075] Another embodiment of the present invention provides a target detection system based on data fusion. For details, see Figure 2 , Figure 2 A schematic diagram of obtaining a loss function in one embodiment of the present invention is shown, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, a target detection method based on data fusion as described in any one of the above is implemented.

[0076] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0077] (1) All agents are used to perform target detection on the historical point cloud dataset. A graph matching matrix is ​​constructed based on the similarity features between all target detection results. The graph matching matrix records the similarity between multiple detection results of different agents, thereby retaining the most original correlation between the detection results.

[0078] (2) Based on the graph matching matrix, each initial vertex feature is obtained. Since the initial feature only records the original node information, when there is a position error in the detection result, the node similarity will decrease, resulting in a weakened correlation between different detection results of the same target. Therefore, in order to reduce the negative impact of the error, the edge similarity information is introduced to correct the vertex features. The first graph convolutional network is used to process the edge weight matrix of the graph matching matrix, the degree matrix of the graph matching matrix and the initial vertex features to obtain updated vertex features. The features of adjacent vertices are learned through the graph convolutional network to update its own features, thereby improving the accuracy of subsequent association fusion analysis and enhancing the robustness of target detection in complex environments, thereby improving the accuracy of target detection;

[0079] (3) A second graph convolutional network is used to perform matching analysis on each updated vertex feature and the remaining updated vertex features to obtain a non-negative score matrix. The target association is converted into a vertex classification problem and solved using a graph convolutional network. The non-negative score matrix is ​​normalized to obtain a double random matrix to meet the one-to-one constraint of the graph matching problem. The elements in the double random matrix represent the matching probability between nodes. A loss function is constructed based on the difference between the double random matrix and the true matching matrix. The graph convolutional neural network model is trained using the loss function. The back-propagation algorithm is used to perform gradient descent to optimize the graph matching results. By minimizing the loss function, the model learns how to generate a prediction probability matrix that is more consistent with the true matching label, thereby gradually improving the matching performance and ensuring the robustness of target detection in complex environments.

[0080] (4) The trained graph convolutional neural network model is used to process all real-time target detection results of the target agent to obtain the detection results of each associated target of the target agent. All associated target detection results are weighted fused to obtain the fused target detection results. The results of multiple detection boxes are integrated through a weighted fusion strategy to further improve the accuracy of target detection.

[0081] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A target detection method based on data fusion, characterized in that: include: All agents are used to perform target detection on the historical point cloud dataset, and a graph matching matrix is ​​constructed based on the similarity features between all target detection results. Acquire each initial vertex feature based on the graph matching matrix; use a first graph convolutional network to process an edge weight matrix of the graph matching matrix, a degree matrix of the graph matching matrix, and the initial vertex features to obtain updated vertex features; Using a second graph convolutional network to perform matching analysis on each of the updated vertex features and the remaining updated vertex features to obtain a non-negative score matrix; normalizing the non-negative score matrix to obtain a double random matrix; Constructing a loss function based on the difference characteristics between the double random matrix and the true matching matrix, and training the graph convolutional neural network model with the loss function; The trained graph convolutional neural network model is used to process all real-time target detection results of the target intelligent agent to obtain each associated target detection result of the target intelligent agent, and all the associated target detection results are weightedly fused to obtain a fused target detection result.

2. The target detection method based on data fusion according to claim 1, characterized in that: The step of constructing a graph matching matrix based on the similarity features between all target detection results obtained includes: Taking each of the target detection results as a node, connecting each node to the remaining nodes, and taking the line between any two nodes as an edge; Constructing a first feature vector of each node based on the label of each target detection result, the shape of the detection box, and the position of the center point of the detection box, and constructing a second feature vector of each edge based on the length and angle of each edge and the label of each corresponding node; Calculating the comprehensive node similarity between the corresponding nodes according to the similarity feature between the first feature vectors of two nodes, and calculating the comprehensive edge line similarity between the corresponding edges according to the similarity feature between the second feature vectors of two edges; The comprehensive node similarity is used as an element at a diagonal position of the graph matching matrix, and the comprehensive edge similarity is used as an element at a non-diagonal position of the graph matching matrix.

3. The target detection method based on data fusion according to claim 2, characterized in that: The calculating the comprehensive node similarity between corresponding nodes according to the similarity feature between the first feature vectors of two nodes comprises: The similarity of the shapes of the detection boxes of the two nodes is used as the first node similarity, the Euclidean distance between the positions of the center points of the detection boxes of the two nodes is inversely proportionally processed to obtain the second node similarity, the similarity of the labels of the two nodes is used as the third node similarity, and the node similarity of the two nodes is calculated based on the linear relationship between the first node similarity, the second node similarity and the third node similarity; The calculating the comprehensive edge line similarity between the corresponding edges according to the similarity feature between the second feature vectors of the two edges comprises: The similarity of the labels of the starting nodes of the two edges is taken as the first edge point similarity, and the similarity of the labels of the ending nodes of the two edges is taken as the second edge point similarity. The third edge similarity is obtained based on the difference characteristics of the lengths of the two edges; the fourth edge similarity is obtained based on the difference characteristics of the angles of the two edges. According to the linear relationship between the first edge point similarity, the second edge point similarity, the third edge similarity and the fourth edge similarity, the comprehensive edge similarity of the two nodes is calculated.

4. The target detection method based on data fusion according to claim 2, characterized in that: The acquiring each initial vertex feature based on the graph matching matrix includes: The elements at all diagonal positions of the graph matching matrix are used as initial vertex features.

5. The target detection method based on data fusion according to claim 1, characterized in that: The first graph convolutional network includes two fully connected layers, and the activation function of the first graph convolutional network is a ReLU activation function; The step of using the first graph convolutional network to process the edge weight matrix of the graph matching matrix, the degree matrix of the graph matching matrix, and the initial vertex features to obtain updated vertex features includes: m (k) =D -1 Wf1(v (k-1) )+f1(v (k-1) ) Among them, m (k) Update the k-th vertex feature, D -1 is the inverse matrix of the degree matrix of the graph matching matrix, W is the edge weight matrix of the graph matching matrix, f1 is the fully connected layer of the first graph convolutional network, v (k-1) is the k-1th initial vertex feature.

6. The target detection method based on data fusion according to claim 1, characterized in that: The second graph convolutional network includes a fully connected layer, and the activation function of the second graph convolutional network is an exponential activation function.

7. The target detection method based on data fusion according to claim 1, characterized in that: The sum of the elements in each row of the double random matrix is ​​1, and the sum of the elements in each column of the double random matrix is ​​1.

8. The target detection method based on data fusion according to claim 1, characterized in that: The loss function is constructed based on the difference characteristics between the double random matrix and the true matching matrix, include: Among them, L is the loss function, n p is the number of updated vertex features, log is the logarithmic function, P i,j is the element value of the i-th row and j-th column in the doubly random matrix, X i,j is the element value of the i-th row and j-th column in the true matching matrix.

9. The target detection method based on data fusion according to claim 1, characterized in that: The weighted fusion of all the associated target detection results to obtain a fused target detection result includes: The classification confidence of each detection frame of the target intelligent body is used as the corresponding weight, and the associated target detection results of each detection frame of the target intelligent body are weighted and summed as the fused target detection result.

10. A target detection system based on data fusion, characterized in that: The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, a target detection method based on data fusion as described in any one of claims 1 to 9 is implemented.

Images