Dynamic target detection method, electronic device and storage medium

Abstract

The present application discloses a method for detecting dynamic targets, an electronic device and a storage medium. The method comprises: obtaining a current video frame; obtaining a preset number of prediction information of the current video frame, wherein the preset number of prediction information of the current video frame are respectively predicted by using a neural network based on information of a preset number of previous video frames of the current video frame, and the neural network is trained using training video frames, and there are no dynamic targets in the training video frames; based on the difference between the information of the current video frame and the prediction information, determining whether there is a dynamic target in the current video frame. In the above manner, the real-time performance of dynamic target detection can be improved.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a dynamic target detection method, electronic device and storage medium. Background Art

[0002] Dynamic target detection is widely used in the field of surveillance. In the field of surveillance, dynamic target detection is needed to detect whether there are dynamic targets in the video frame. If there are dynamic targets in the video frame, the video frame is considered abnormal and an alarm is required so that relevant personnel can respond in time.

[0003] Existing dynamic target detection technologies mainly include optical flow method, frame difference method and background difference method, etc. The basic principle of existing dynamic target detection technology is to subtract two adjacent video frames in a video frame sequence to find the video frame that has changed in the video frame sequence, or to extract the moving target in the current video frame by subtracting the current video frame from the background model.

[0004] However, the existing dynamic target detection methods are not real-time enough, which results in the inability of relevant personnel to respond in time, causing adverse consequences. Summary of the invention

[0005] The present application provides a dynamic target detection method, an electronic device and a storage medium, which can solve the problem that the existing image downsampling method is not highly applicable.

[0006] In order to solve the above technical problems, a technical solution adopted by the present application is: to provide a dynamic target detection method. The method includes: obtaining a current video frame; obtaining a preset number of prediction information of the current video frame, wherein the preset number of prediction information of the current video frame is respectively predicted by using a neural network based on information of a preset number of previous video frames of the current video frame, and the neural network is trained using training video frames, and there is no dynamic target in the training video frame; based on the difference between the information of the current video frame and the prediction information, determining whether there is a dynamic target in the current video frame.

[0007] To solve the above technical problems, another technical solution adopted in the present application is: to provide an electronic device, which includes a processor and a memory connected to the processor, wherein the memory stores program instructions; the processor is used to execute the program instructions stored in the memory to implement the above method.

[0008] In order to solve the above technical problem, another technical solution adopted by the present application is: providing a storage medium storing program instructions, which can implement the above method when executed.

[0009] Through the above method, since the neural network is trained using training video frames without dynamic targets, the prediction accuracy of video frames without dynamic targets is higher than the prediction accuracy of video frames with dynamic targets, and the prediction is based on the condition that there are no dynamic targets. When a dynamic target appears, there will inevitably be a large difference between the information of the video frame at that moment and the predicted information. Therefore, the present application can determine whether there is a dynamic target in the current video frame based on the difference between the information of the current video frame and the predicted information of the current video frame obtained using the neural network.

[0010] Moreover, compared with the method of directly using a neural network to judge the current video frame (this method requires a large number of labeled video frames, but cannot cover all types of dynamic targets, and the network model is complex, training is difficult, and the network forward propagation is time-consuming, affecting the real-time detection. In addition, in the same monitoring scenario, there is the problem that the neural network parameters need to be updated due to the introduction of new labels), the present application only needs to use a neural network to obtain the prediction information of the current video frame, and then determine whether there is a dynamic target in the current video frame based on the difference between the information of the current video frame and the prediction information. This method can reduce the delay caused by the direct forward propagation of the neural network for dynamic target judgment and detection, and improve the real-time detection of dynamic targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a structural diagram of a dynamic target detection system of the present application;

[0012] Figure 2 It is another structural schematic diagram of the dynamic target detection system of the present application;

[0013] Figure 3 It is a flowchart of the first embodiment of the dynamic target detection method of the present application;

[0014] Figure 4 It is a flow chart of the second embodiment of the dynamic target detection method of the present application;

[0015] Figure 5 yes Figure 4 The specific process diagram of S22;

[0016] Figure 6 It is a flowchart of the third embodiment of the dynamic target detection method of the present application;

[0017] Figure 7 yes Figure 6 The specific process diagram of S32;

[0018] Figure 8 is a flowchart of a fourth embodiment of the dynamic target detection method of the present application;

[0019] Fig. 9It is a structural schematic diagram of an embodiment of the electronic device of the present application;

[0020] Fig.10 It is a structural diagram of an embodiment of a storage medium of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] The terms "first", "second", and "third" in this application are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first", "second", and "third" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0023] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments without conflict.

[0024] Before introducing the dynamic target detection method provided by the present application, the application scenario of the method is described first. The dynamic target detection method provided by the present application can be implemented by relying on a dynamic target detection system.

[0025] See also Figure 1 In an application scenario, a dynamic target detection system may include a sensor and a detection device, and the detection device has detection capabilities. The sensor is arranged in the detection area, and the sensor can collect video frames of the detection area and send the video frames to the detection device through the communication system, so that the detection device detects whether there is a dynamic target in the video frame. When the detection result is that there is a dynamic target, the detection device can indicate each linkage item to alarm.

[0026] See also Figure 2In another application scenario, the dynamic target detection system may include sensors, cloud servers and detection devices, and the detection device itself does not have detection capabilities. The sensor is deployed in the detection area, and the sensor can collect video frames of the detection area and send the video frames to the cloud server and the detection device respectively through the communication system. The detection device can display and store video frames in real time, and the cloud server can detect whether there are dynamic targets in the video frames. If there are dynamic targets, the cloud server can send an alarm signal and video frames to the detection device through the communication system, so that the detection device indicates each linkage item alarm.

[0027] The type of sensors involved in the above application scenarios depends on the type of video frame information that needs to be collected. For specific instructions, please refer to the description later in this application.

[0028] It should be noted that the above application scenarios are only examples and are not meant to limit the application scenarios of the present application.

[0029] Figure 3 is a flow chart of the first embodiment of the dynamic target detection method of the present application. It should be noted that if there are substantially the same results, this embodiment does not Figure 3 The process sequence shown is limited. Figure 3 As shown, this embodiment may include:

[0030] S11: Get the current video frame.

[0031] The video frames mentioned in this application can be obtained through sensors. The video frames may include information such as pixel data, timestamps, etc. of the image (in YUV, YCbCr, etc. format) obtained by the camera sensor, or, in addition to the above information, the video frames may also include brightness information of the environment where the camera sensor is located, which can be used to reflect the light intensity of the detection area, and the brightness information of the environment where the camera sensor is located can be obtained through the light sensor. S12: Obtain a preset number of prediction information of the current video frame.

[0032] Among them, the preset number of prediction information of the current video frame are respectively predicted by using a neural network based on the information of a preset number of previous video frames of the current video frame. The neural network is trained using training video frames, and there are no dynamic targets in the training video frames.

[0033] The information of the video frame referred to in this application may include the brightness information of the video frame (i.e., the brightness component Y in the image pixel data). Alternatively, the information of the video frame referred to in this application may include the brightness information and illumination information of the video frame. Among them, the illumination information of the video frame may be the illumination information of the detection area to which the video frame belongs, i.e., the brightness information of the environment where the camera sensor is located as mentioned above.

[0034] The method of predicting the current video frame x(N) to obtain N pieces of prediction information is described below. When video frame x(0) is obtained, the information of video frame x(0) is predicted N frames backward to obtain prediction information (x(0, 1) to x(0, ​​N)) of N subsequent videos (x(1) to x(N)) of video frame x(0); when video frame x(1) is obtained, the information of video frame x(1) is predicted N frames backward to obtain prediction information (x(1, 2) to x(1, N+1)) of N subsequent videos (x(2) to x(N+1)) of video frame x(1); ...; and so on. When video frame x(N-1) is obtained, the information of video frame x(N-1) is predicted N frames backward to obtain prediction information (x(N-1, N) to x(N-1, 2N-1)) of N subsequent videos (x(N) to x(2N-1)) of video frame x(N-1).

[0035] Among them, x(0, ​​N) is the prediction information of x(N) based on x(0), x(1, N) is the prediction information of x(N) based on x(1), and x(N-1, N) is the prediction information of x(N) based on x(N-1). Therefore, the prediction information sequence of the current video frame x(N) is {x(0, N), x(1, N), ..., x(N-1, N)}.

[0036] The neural network used in this application may be SAE-LSTM, or other neural networks that can predict the current video frame. For a detailed description of neural network training, please refer to the following embodiments.

[0037] S13: Determine whether there is a dynamic target in the current video frame based on the difference between the information of the current video frame and the predicted information.

[0038] It is understandable that, since the neural network is trained based on training video frames without dynamic targets, the accuracy of the prediction information of video frames with dynamic targets obtained by the neural network is lower than the accuracy of the prediction information of video frames without dynamic targets, and the prediction is based on the condition of no dynamic targets. When a dynamic target appears, there will inevitably be a large difference between the information of the video frame at that moment and the prediction information. Therefore, it is possible to determine whether there is a dynamic target in the current video frame based on the difference between the information of the current video frame and the prediction information.

[0039] In a specific implementation, whether there is a dynamic target in the current video frame can be determined based on the size of the difference between the information of the current video frame and the predicted information. For example, a preset threshold can be set. If the size of the difference between the information of the current video frame and the predicted information exceeds the preset threshold, it is determined that there is a dynamic target in the current video frame; otherwise, it is determined that there is no dynamic target in the current video frame. The size of the difference between the information of the current video frame and the predicted information can be specifically reflected in the size of the Euclidean distance (the sum of squares of prediction errors) between the information of the current video frame and the predicted information, the size of the absolute distance (the sum of absolute values ​​of prediction errors), etc.

[0040] In order to improve the accuracy of the determination result, in another specific implementation, whether there is a dynamic target in the current video frame can be determined based on the difference distribution between the information of the current video frame and each prediction information. For example, when it is known that there is no dynamic target, the difference distribution between the information of the video frame and the prediction information (reference distribution) can be used to determine whether there is a dynamic target in the current video frame based on the difference between the difference distribution between the information of the current video frame and each prediction information and the reference distribution.

[0041] The reference distribution may represent the difference between the information of the video frame when there is no dynamic target and the predicted information of the video frame obtained by using the neural network. It may be the difference between the information of the training video frame and the predicted information of the training video frame obtained by the last iteration training process of the neural network. It may also be set in other ways.

[0042] Through the implementation of this embodiment, since the neural network obtained by training with training video frames without dynamic targets has a higher prediction accuracy for video frames without dynamic targets than for video frames with dynamic targets, and the prediction is performed based on the condition that there are no dynamic targets, when a dynamic target appears, there will inevitably be a large difference between the video frame at that moment and the previous video frame at that moment. Therefore, the present application can determine whether there is a dynamic target in the current video frame based on the difference between the information of the current video frame and the prediction information of the current video frame obtained by using the neural network.

[0043] Moreover, compared with the method of directly using a neural network to judge the current video frame (this method requires a large number of labeled video frames, but cannot cover all types of dynamic targets, and the network model is complex, difficult to train, and the network forward propagation takes a long time, affecting the real-time detection. In addition, in the same monitoring scene, there is a problem that the neural network parameters need to be updated due to the introduction of new labels), the present application only needs to use a neural network to obtain the prediction information of the current video frame, and then determine whether there is a dynamic target in the current video frame based on the difference between the information of the current video frame and the prediction information. This method can reduce the delay caused by the direct forward propagation of the neural network for dynamic target judgment and detection, and improve the real-time detection of dynamic targets. In addition, in the above embodiment, after obtaining the current video frame, the neural network is also used to predict the preset number of subsequent video frames of the current video frame based on the information of the current video frame, and the prediction information of the preset number of subsequent video frames of the current video frame is obtained for subsequent use.

[0044] Figure 4 is a flow chart of the second embodiment of the dynamic target detection method of the present application. It should be noted that if there are substantially the same results, this embodiment does not Figure 4 The process sequence shown is limited. In this embodiment, the information of the current video frame includes brightness information and illumination information, the prediction information of the current video frame includes predicted brightness information and predicted illumination information, and there is no abnormal illumination information in the training video frame. S21 is a further extension of the above S12, and S22 is a further extension of the above S13. Figure 4 As shown, this embodiment may include:

[0045] S21: Obtain a preset number of predicted brightness information and a preset number of predicted illumination information of the current video frame.

[0046] Among them, a preset number of predicted brightness information of the current video frame are respectively predicted by using a neural network based on the brightness information of a preset number of previous video frames of the current video frame, and a preset number of predicted illumination information of the current video frame are respectively predicted by using a neural network based on the illumination information of a preset number of previous video frames of the current video frame.

[0047] For the method of obtaining the predicted brightness information and predicted illumination information of the current video frame in this step, please refer to the detailed description of obtaining the information of the current video frame in the previous embodiment, which will not be repeated here.

[0048] S22: Determine whether there is a dynamic target in the current video frame based on the difference between the brightness information of the current video frame and the predicted brightness information, and the difference between the illumination information of the current video frame and the predicted illumination information. As mentioned above, whether there is a dynamic target in the current video frame can be determined based on the difference size or the difference distribution. The following specifically describes S22 by determining whether there is a dynamic target in the current video frame based on the difference distribution. Figure 5 , S22 may include the following sub-steps:

[0049] S221: Obtain a preset number of first prediction error values ​​between brightness information of the current video frame and predicted brightness information, and obtain a preset number of second prediction error values ​​between illumination information of the current video frame and predicted illumination information.

[0050] For example, the preset number is 5, the brightness information of the current video frame x(5) is x(5a), and the predicted brightness information sequence is {x(50a), x(51a), x(52a), x(53a), x(54a)}, then the first prediction error value sequence of the current video frame x(5) is {|x(5a)-x(50a)|, |x(5a)-x(51a)|, |x(5a)-x(52a)|, |x(5a)-x(53a)|, |x(5a)-x(54a)|}.

[0051] The illumination information of the current video frame x(5) is x(5b) and the predicted illumination information sequence is {x(50b), x(51b), x(52b), x(53b), x(54b)}. Then the second prediction error value sequence of the current video frame x(5) is {|x(5b)-x(50b)|, |x(5b)-x(51b)|, |x(5b)-x(52b)|, |x(5b)-x(53b)|, |x(5b)-x(54b)|}.

[0052] Each first prediction error value may be used to represent the difference between the brightness information of the current video frame and the corresponding predicted brightness information, and each second prediction error value may be used to represent the difference between the illumination information of the current video frame and the corresponding predicted illumination information.

[0053] S222: Determine whether there is a dynamic target in the current video frame based on the distribution of the first prediction error value and the distribution of the second prediction error value.

[0054] The distribution referred to in this application is a probability distribution, such as Gaussian distribution and the like.

[0055] Please refer to the following embodiments for a detailed description of this step, which will not be repeated here.

[0056] Through the implementation of this embodiment, since there is no abnormal illumination information in the training video frame used to train the neural network, the prediction accuracy of the trained neural network for the video frame without abnormal illumination information is higher than the prediction accuracy for the video frame with abnormal illumination information, and the prediction of the illumination information is based on the condition of no abnormal illumination information. When the abnormal illumination information appears, a large difference will inevitably occur between the video frame at that moment and the previous video frame at that moment. Therefore, based on the difference between the illumination information of the current video frame and the predicted illumination information, the present application can determine whether there is abnormal illumination information in the current video frame, that is, it can determine whether the illumination information will interfere with the process of determining whether there is a dynamic target in the current video frame by brightness information. Therefore, in the present application, the difference between the brightness information of the current video frame and the predicted brightness information, and the difference between the illumination information of the current video frame and the predicted illumination information are combined to jointly determine whether there is a dynamic target in the current video frame, which can make the determination result more accurate.

[0057] Figure 6 is a flow chart of the third embodiment of the dynamic target detection method of the present application. It should be noted that if there are substantially the same results, this embodiment does not Figure 6 The process sequence shown is limited. This embodiment is a further extension of S222. Figure 6 As shown, this embodiment may include:

[0058] S31: Compare the distribution of the first prediction error value with the first benchmark distribution to obtain a first difference score, and compare the distribution of the second prediction error value with the second benchmark distribution to obtain a second difference score.

[0059] The first reference distribution may represent the distribution of the first prediction error value when there is no dynamic target, and the second reference distribution may represent the distribution of the second prediction error value when there is no dynamic target.

[0060] Referring to the method of obtaining the benchmark distribution mentioned above, it can be known that the first benchmark distribution can be the difference distribution between the brightness information of the training video frame and the predicted brightness information of the training video frame obtained by the last iterative training process of the neural network, and the second benchmark distribution can be the difference distribution between the illumination information of the training video frame and the predicted illumination information of the training video frame obtained by the last iterative training process of the neural network.

[0061] The first difference algorithm may be used to compare the distribution of the first prediction error value with the first benchmark distribution to obtain a first difference score, and the first difference algorithm may be used to compare the distribution of the second prediction error value with the second benchmark distribution to obtain a second difference score.

[0062] And / or, a second difference algorithm may be used to compare the distribution of the first prediction error value with the first benchmark distribution to obtain a first difference score, and a second difference algorithm may be used to compare the distribution of the second prediction error value with the second benchmark distribution to obtain a second difference score.

[0063] The accuracy of the first difference algorithm is higher than that of the second difference algorithm. It is understandable that, compared with the algorithm with low accuracy, the algorithm with high accuracy takes longer time and thus has a slower detection speed.

[0064] Optionally, the first difference algorithm is a generalized log-likelihood ratio (GLLR) test and a cumulative algorithm, and the second difference algorithm is a relative entropy (KL divergence) test.

[0065] S32: Determine whether there is a dynamic target in the current video frame based on the difference between the first difference score and the second difference score.

[0066] It can be understood that whether there is a dynamic target in the current video frame is determined based on the difference between the first difference score and the second difference score, that is, the second difference score obtained by using the illumination information compensates the first difference score obtained by using the brightness information, thereby reducing the error of the determination result caused by the illumination change.

[0067] See also Figure 7 , S32 may include the following sub-steps:

[0068] S321: Obtain a difference between the first difference score and the second difference score.

[0069] S322: Compare the difference with a preset threshold to obtain a comparison result.

[0070] For example, the first difference algorithm is a generalized log-likelihood ratio test and a cumulative algorithm, and the second difference algorithm is a relative entropy test. The preset threshold corresponding to the first difference algorithm can be set according to actual needs, and the preset threshold corresponding to the second difference algorithm has a value range of γ times the preset threshold corresponding to the first difference algorithm, and γ takes a value of [0,1).

[0071] S323: Determine whether there is a dynamic target in the current video frame based on the comparison result.

[0072] The present embodiment is described below in the form of an example.

[0073] S31 is performed using a generalized log-likelihood ratio test and an accumulation algorithm to obtain a first comparison result. S31 is performed using a relative entropy test to obtain a second comparison result.

[0074] The speed of obtaining the second comparison result is faster than the speed of obtaining the first comparison result, and the accuracy of the first comparison result is higher than that of the second comparison result. When the second comparison result is obtained first, indicating the existence of a dynamic target, the buzzer associated with the detection device will sound a low-frequency alarm; when the first comparison result is obtained later, indicating the existence of a dynamic target, the buzzer associated with the detection device will sound a high-frequency alarm. Among them, when the buzzer sounds a low-frequency alarm, it means that the alarm level is low, and relevant personnel can choose whether to respond. When the buzzer sounds a high-frequency alarm, it means that the alarm level is high, and relevant personnel need to respond in a timely manner.

[0075] Through the implementation of this embodiment, when determining whether there is a dynamic target in the current video frame, the application takes into account the impact of lighting changes on the brightness information of the current video frame, which can improve the accuracy of the determination result.

[0076] The training process of the neural network of the present application is described below. Figure 8 is a flow chart of the third embodiment of the dynamic target detection method of the present application. It should be noted that if there are substantially the same results, this embodiment does not Figure 8 The process sequence shown is limited. Figure 8 As shown, this embodiment may include:

[0077] S41: Obtain training video frames.

[0078] The training video frames can be video frames newly collected by the sensor or video frames collected historically.

[0079] It can be understood that, in the case where the information of the video frame in the previous embodiment includes brightness information, there is no dynamic target in the training video frame used to train the neural network; in the case where the information of the video frame in the previous embodiment includes brightness information and illumination information, there is no dynamic target and no abnormal illumination information in the training video frame used to train the neural network. This embodiment takes the latter as an example to illustrate the process of training the neural network.

[0080] Since there may be dynamic targets and / or abnormal illumination information in the video frames acquired by the sensor, the acquired video frames may be preprocessed such as denoising and normalization to select video frames in which there are no dynamic targets and no abnormal illumination information as training video frames.

[0081] S42: using a neural network to obtain a preset number of predicted brightness information of the training video frame based on the brightness information of a preset number of previous training video frames of the training video frame, and using a neural network to obtain a preset number of predicted brightness information of the training video frame based on the illumination information of a preset number of previous training video frames of the training video frame.

[0082] Please refer to the previous embodiments for detailed description of this embodiment, which will not be repeated here.

[0083] S43: Adjusting the parameters of the neural network based on the difference between the brightness information of the training video frame and the predicted brightness information, and the difference between the illumination information of the training video frame and the predicted illumination information.

[0084] For example, the sum of squares of the prediction error values ​​and the sparse penalty can be used together to construct a loss function to update the neural network parameters until the loss function value is less than a preset loss value or the number of iterations reaches a preset number of iterations.

[0085] In addition, the difference distribution between the information of the training video frame and the predicted information of the training video frame obtained by the last iterative training process of the neural network can be used as the reference distribution corresponding to the video frame when there is no dynamic target in the subsequent use process.

[0086] Fig. 9 Schematic diagram of the structure of an embodiment of the electronic device of the present application. Fig. 9 As shown, the electronic device includes a processor 51 and a memory 52 coupled to the processor 51 .

[0087] The memory 52 stores program instructions for implementing the method of any of the above embodiments; the processor 51 is used to execute the program instructions stored in the memory 52 to implement the steps of the above method embodiments. The processor 51 may also be referred to as a CPU (Central Processing Unit). The processor 51 may be an integrated circuit chip with signal processing capabilities. The processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0088] Fig.10 Schematic diagram of the structure of a storage medium of the present application. Fig.10As shown, the computer-readable storage medium 60 of the embodiment of the present application stores program instructions 61, and the program instructions 61 are executed to implement the method provided in the above embodiment of the present application. Among them, the program instructions 61 can form a program file and be stored in the above computer-readable storage medium 60 in the form of a software product, so that a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) executes all or part of the steps of each implementation method of the present application. The aforementioned computer-readable storage medium 60 includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0089] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0090] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the description and drawings of this application, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A dynamic target detection method, characterized in that: include: Get the current video frame; Obtaining a preset number of prediction information of the current video frame, wherein the preset number of prediction information of the current video frame are respectively predicted by using a neural network based on information of a preset number of previous video frames of the current video frame, and the neural network is trained using training video frames, and the dynamic target does not exist in the training video frames; Determining whether the dynamic target exists in the current video frame based on a difference between the information of the current video frame and the prediction information; Wherein, determining whether the dynamic target exists in the current video frame based on the difference between the information of the current video frame and the predicted information includes: The information of the current video frame includes brightness information and illumination information, and the predicted information of the current video frame includes predicted brightness information and predicted illumination information. A preset number of first prediction error values ​​between the brightness information of the current video frame and the predicted brightness information are obtained, and a preset number of second prediction error values ​​between the illumination information of the current video frame and the predicted illumination information are obtained; the distribution of the first prediction error value is compared with the first benchmark distribution to obtain a first difference score; the distribution of the second prediction error value is compared with the second benchmark distribution to obtain a second difference score; the difference between the first difference score and the second difference score is obtained; the difference is compared with a preset threshold to obtain a comparison result; based on the comparison result, it is determined whether the dynamic target exists in the current video frame; the first benchmark distribution represents the difference distribution between the brightness information of the video frame when there is no dynamic target and the predicted brightness information of the video frame obtained by using a neural network, and the second benchmark distribution represents the difference distribution between the illumination information of the video frame when there is no dynamic target and the predicted illumination information of the video frame obtained by using a neural network.

2. The method according to claim 1, characterized in that There is no abnormal illumination information in the training video frame, and obtaining a preset number of prediction information of the current video frame includes: Obtain a preset number of predicted brightness information and a preset number of predicted illumination information of the current video frame, wherein the preset number of predicted brightness information of the current video frame are respectively predicted by the neural network based on the brightness information of the preset number of previous video frames of the current video frame, and the preset number of predicted illumination information of the current video frame are respectively predicted by the neural network based on the illumination information of the preset number of previous video frames of the current video frame.

3. The method according to claim 1, characterized in that The step of comparing the distribution of the first prediction error value with the first reference distribution to obtain a first difference score, and comparing the distribution of the second prediction error value with the second reference distribution to obtain a second difference score comprises: Comparing the distribution of the first prediction error value with the first benchmark distribution using a first difference algorithm to obtain the first difference score, and comparing the distribution of the second prediction error value with the second benchmark distribution using the first difference algorithm to obtain the second difference score; and / or, The distribution of the first prediction error value is compared with the first benchmark distribution using a second difference algorithm to obtain the first difference score, and the distribution of the second prediction error value is compared with the second benchmark distribution using the second difference algorithm to obtain the second difference score.

4. The method according to claim 3, characterized in that The accuracy of the first difference algorithm is higher than the accuracy of the second difference algorithm; And / or, the first difference algorithm is a generalized log-likelihood ratio test and a cumulative algorithm, and the second difference algorithm is a relative entropy test.

5. The method according to claim 1, characterized in that: The training of the neural network includes: Acquire the training video frame; Using the neural network based on brightness information of a preset number of previous training video frames of the training video frame, obtaining a preset number of predicted brightness information of the training video frame; using the neural network based on illumination information of a preset number of previous training video frames of the training video frame, obtaining a preset number of predicted illumination information of the training video frame; Based on the difference between the brightness information of the training video frame and the predicted brightness information, and the difference between the illumination information of the training video frame and the predicted illumination information, the parameters of the neural network are adjusted.

6. The method according to claim 5, characterized in that Before comparing the distribution of the first prediction error value with the first benchmark distribution to obtain a first difference score and comparing the distribution of the second prediction error value with the second benchmark distribution to obtain a second difference score, the method includes: The difference distribution between the brightness information of the training video frame and the predicted brightness information of the training video frame obtained by the last iterative training process of the neural network is used as the first benchmark distribution, and the difference distribution between the illumination information of the training video frame and the predicted illumination information of the training video frame obtained by the last iterative training process of the neural network is used as the second benchmark distribution.

7. The method according to claim 1, characterized in that After obtaining the current video frame, the method further includes: The neural network is used to predict a preset number of subsequent video frames of the current video frame based on information of the current video frame to obtain prediction information of the preset number of subsequent video frames of the current video frame.

8. An electronic device, characterized in that: The method comprises a processor and a memory connected to the processor, wherein: The memory stores program instructions; The processor is configured to execute the program instructions stored in the memory to implement the method according to any one of claims 1 to 7.

9. A storage medium, characterized in that: The storage medium stores program instructions, and when the program instructions are executed, the method according to any one of claims 1 to 7 is implemented.

Images