Data processing device and data processing method

Abstract

To provide a technology for determining analogies between environments that are the targets of recognition processing.SOLUTION: A data processing device 20 is provided with a comparison unit 33 for determining analogies between a first processing result, obtained by recognizing events occurring in a reference environment through predetermined recognition processing, and a second processing result, obtained by recognizing events occurring in a target environment through recognition processing, and a determination unit 34 for determining whether the reference environment and the target environment are similar based on the analogy between the first comparison processing and the second comparison processing.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a data processing device and a data processing method. [Background technology]

[0002] Currently, recognition systems are being used that recognize events occurring in an observed environment based on the output signals of sensors installed in the environment. When applying such a recognition system to an actual real environment, it is necessary to create a virtual environment that mimics the real environment before actually applying it, execute a recognition process in the virtual environment, evaluate the recognition performance, and then adjust the processing parameters used in the recognition process to achieve the desired recognition performance.

[0003] In the past, the recognition performance of recognition processing was evaluated and verified by running the recognition processing in a physically constructed virtual environment. In recent years, however, a method has been adopted in which a virtual environment is constructed using CG (computer graphics) and the recognition processing is executed in that virtual environment, thereby reducing the cost of constructing the virtual environment. For example, Patent Document 1 describes a technique for improving recognition accuracy by re-learning training images generated using CG. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2016 / 151618 Brochure Summary of the Invention [Problem to be solved by the invention]

[0005] However, as the number of real environments to which the recognition system is applied increases, the operational burden increases if a virtual environment is individually constructed for each real environment. In this regard, if there is a known environment in which the processing parameters used in the recognition process have been adjusted, and this known environment is similar to the real environment, the processing parameters adjusted for this known environment can be utilized to adjust the processing parameters used in the recognition process in the real environment. The present invention has been made in consideration of the above-mentioned problems, and has as its object to provide a technique for determining similarity between environments that are the subject of recognition processing. [Means for solving the problem]

[0006] A data processing device according to one embodiment of the present invention comprises a comparison unit that determines the similarity between a first processing result obtained by recognizing an event occurring in a reference environment through a predetermined recognition process and a second processing result obtained by recognizing an event occurring in a target environment through a recognition process, and a determination unit that determines whether the reference environment and the target environment are similar based on the similarity between the first processing result and the second processing result.

[0007] A data processing device according to another aspect of the present invention includes a memory unit that stores a third environmental parameter, which is an environmental parameter indicating a reference environment, and a base parameter, which is a processing parameter used in a recognition process to recognize an event that occurs in the reference environment; a judgment unit that determines whether the other environment is similar to the reference environment based on an analogy between the second environmental parameter, which is an environmental parameter indicating an environment other than the reference environment, and the third environmental parameter; and a processing parameter determination unit that, when it is determined that the other environment is similar to the reference environment, determines the base parameter as the processing parameter to be used in the recognition process to recognize an event that occurs in the other environment. [Effects of the Invention]

[0008] According to the present invention, a technique for determining similarity between environments that are the subject of recognition processing can be provided. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a schematic diagram illustrating an example of the overall configuration of a recognition system that recognizes events that occur in a real environment. [Figure 2] 1A to 1C are schematic diagrams illustrating an example of a data processing method according to an embodiment. [Figure 3] FIG. 1 is a schematic diagram illustrating an example of a hardware configuration of a data processing device according to an embodiment. [Figure 4] 1 is a block diagram illustrating an example of a functional configuration of a data processing apparatus according to an embodiment. [Figure 5] 10A and 10B are schematic diagrams illustrating an example of an intermediate result of a person tracking process. [Figure 6] 10(a) to 10(e) are schematic diagrams of a reference environment, a target environment, and a similarity judgment. [Figure 7] 10A and 10B are schematic diagrams illustrating an example of a method for adjusting specialized parameters of base parameters. [Figure 8] 1 is a flowchart illustrating an example of a data processing method according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the embodiments of the present invention shown below are merely examples of devices and methods for embodying the technical concept of the present invention, and the technical concept of the present invention does not limit the structure, arrangement, etc. of the components to those described below. The technical concept of the present invention can be modified in various ways within the technical scope defined by the claims.

[0011] A data processing device according to an embodiment of the present invention accepts data acquired in a specific environment as input data and determines processing parameters to be used in recognition processing that recognizes events that occur in this environment based on the input data. For example, the recognition process may be a tracking process or a detection process of an object in an input image that is input data (e.g., person detection, head detection, abnormality detection, intruder detection, etc.), or a matching process of an object in an input image (e.g., face recognition, etc.). Note that the input data is not limited to an image, and the recognition process may be a recognition process based on input voice that is input data (e.g., voice recognition, etc.), or a recognition process based on power consumption data (e.g., abnormality detection, etc.). In the following description, an example of a data determination device that determines processing parameters for recognition processing based on an input image will be described as an example of a data processing device according to an embodiment of the present invention.

[0012] Furthermore, the recognition logic used in the recognition process may be, for example, a learning model (such as a neural network) constructed by machine learning, or rule-based artificial intelligence, or a combination of these. Furthermore, the processing parameters used in the recognition process may be thresholds to be compared with a detection score indicating the detection of an event or object to be detected in the recognition process, or may be coefficients of a calculation function for calculating the detection score.

[0013] FIG. 1 is a schematic diagram showing an example of the overall configuration of a recognition system 1 in which processing parameters are determined by a data processing device according to an embodiment. The recognition system 1 recognizes events occurring in a specific environment 2 (e.g., a monitored area). When applying the recognition system 1 to the actual environment 2, it is necessary to adjust the processing parameters used in the recognition process so as to achieve the desired recognition performance. In the following description, the environment 2 to which the recognition system 1 is actually applied may be referred to as the "real environment."

[0014] The recognition system 1 includes a sensor 3 for monitoring the real environment 2, and a recognition device 10 that recognizes events occurring in the real environment 2 using the output value of the sensor 3 as input data. In the following explanation, a case where the sensor 3 is a camera that captures images of the real environment 2 will be exemplified. The recognition device 10 includes an input unit 12, a communication unit 13, a storage unit 14, a control unit 15, and an output unit 16. Of these, the storage unit 14 and the control unit 15 can be realized by a so-called computer, and the input unit 12, the communication unit 13, and the output unit 16 can be realized as peripheral devices of the computer.

[0015] The input unit 12 includes a user interface such as a keyboard, a mouse, etc. that is operated by a user to input data, etc. The input unit 12 is connected to the control unit 15, converts the user's operation into an operation signal, and outputs it to the control unit 15. The input unit 12 may also include a DVD (Digital Versatile Disc) drive and a USB (Universal Serial Bus) interface. The input unit 12 inputs data from these storage media to the control unit 15 as a file.

[0016] The communication unit 13 includes a network interface or the like that transmits and receives data between the recognition device 10 and an external device (for example, a data processing device 20 described later) via wired or wireless communication. The storage unit 14 is a memory device such as a ROM (Read Only Memory) or a RAM (Random Access Memory), and stores various programs and various data. The control unit 15 is a controller that acquires input images captured by the sensor 3 and recognizes events occurring in the real environment 2 based on the input images, and is composed of arithmetic devices (processors) such as a CPU (Central Processing Unit), a DSP (Digital Signal Processor), an MCU (Micro Control Unit), and a GPU (Graphics Processing Unit).

[0017] The control unit 15 is connected to the storage unit 14, and operates as various processing units by reading and executing computer programs from the storage unit 14, and stores various data in the storage unit 14 and reads out the data. The control unit 15 is also connected to the input unit 12 to accept operation inputs to the recognition device 10 by the user.

[0018] The control unit 15 executes a recognition process to recognize an event that has occurred in the real environment 2 based on an image of the real environment 2 captured by the camera 3. The recognition process by the control unit 15 may be, for example, a tracking process or a detection process (for example, person detection, head detection, abnormality detection, intruder detection, etc.) of an object (for example, a person) in the real environment 2, or a matching process (for example, face recognition, etc.). The recognition logic used in the control unit 15 may be, for example, a learning model (such as a neural network) constructed by machine learning, or may be a rule-based artificial intelligence, or may be a combination of these.

[0019] The control unit 15 is connected to the output unit 16, and outputs the processing results of the control unit 15 to the output unit 16. The control unit 15 is also connected to the communication unit 13, and transmits and receives various data to and from external devices via the communication unit 13. For example, the control unit 15 may acquire processing parameters used in the recognition process by the control unit 15 from a data processing device 20 described below.

[0020] Next, an example of the data processing method according to the embodiment will be outlined below: In the data processing method according to the embodiment, processing parameters used in the recognition process by the control unit 15 are determined. As described above, conventionally, when adjusting the processing parameters used in the recognition process, a virtual environment simulating the real environment 2 is constructed, the recognition process is performed in the virtual environment, the recognition performance is evaluated, and then the parameters are adjusted to achieve the desired recognition performance.

[0021] However, as the number of real environments 2 to which the recognition system 1 is applied increases, the operational load increases if a virtual environment is individually constructed for each real environment 2. Therefore, in the data processing method of the embodiment, a reference environment similar to the real environment 2 is identified from among known reference environments for which processing parameters have been adjusted, and the processing parameters adjusted for the reference environment identified as similar to the real environment 2 are utilized to adjust the processing parameters used in the recognition processing in the real environment 2.

[0022] Here, as factors for determining whether the real environment 2 is similar to the reference environment, conditions that affect the performance of the recognition processing by the control unit 15 (hereinafter sometimes referred to as "environment parameters") can be used among various parameters that indicate the characteristics of the environment. Therefore, in the data processing method of the embodiment, an environmental range is set for each reference environment, which is the range of environmental parameters of environments similar to the reference environment, and a real environment 2 having environmental parameters with values ​​within the environmental range is determined to be similar to the reference environment.

[0023] For example, environmental parameters relating to the space in the real environment 2 may include the size of the real environment 2, its use (entrance / exit, office, etc.), external light, climate, background texture, lighting (brightness, color), how shadows are cast by objects, the arrangement (position and shape) of objects such as fixtures in the real environment 2, the type and arrangement of doors and windows, and the materials (material, reflectance, diffusion coefficient) of objects present in the real environment 2. Furthermore, if the real environment 2 is outdoors, the position, shape, color, and movement of plants may also be included. Furthermore, for example, environmental parameters relating to the shooting conditions in the real environment 2 may include the installation height of the cameras 3, the number of cameras, the frame rate, the angle of view, the resolution, the lens model, distortion, gain, and white balance. Furthermore, for example, environmental parameters relating to the detection target in the real environment 2 may include the number of people, height, body type, behavior, posture, and belongings of people who are the targets of recognition processing in the real environment 2.

[0024] In the data processing method according to the embodiment, in order to set an environment range, a set (hereinafter referred to as a "similar environment set") is formed to include a plurality of environments similar to a reference environment. The similar environment set may include the reference environment itself. Then, the environment range is calculated based on the environmental parameters of the environments belonging to the similar environment set. For example, the distribution range of the environmental parameters of the environments belonging to the similar environment set may be calculated as the environment range. In order to form a similar environment set by collecting environments similar to the reference environment, it is necessary to determine whether multiple candidate environments that form the similar environment set are similar to the reference environment, and to classify the multiple environments according to their similarity to the reference environment. In the following description, the candidate environments that form the similar environment set may be referred to as "target environments."

[0025] In order to determine the similarity between the reference environment and the target environment, the data processing method of the embodiment determines the similarity between the processing results obtained by executing a predetermined recognition process in these environments. This makes it possible to determine whether the reference environment and the target environment are similar to each other in terms of the recognition process of the recognition system, and to form a set of environments that are similar in terms of the recognition process as a similar environment set. As a result, the environmental range of the set of environments that are similar in terms of the recognition process can be calculated.

[0026] 2(a) to 2(c) are schematic diagrams illustrating the above-mentioned outline of an example of a data processing method according to an embodiment. In FIGS. 2(a) to 2(c), circular plots ER1 and ER2 represent reference environments, square plot ET represents a target environment, and triangular plot EA represents a real environment 2. Note that the layout of each environment in FIGS. 2(a) to 2(c) does not represent the layout of these environments in real physical space, but rather represents a schematic representation of the position determined by the environmental parameters of these environments in a parameter space with the environmental parameters as axes.

[0027] In Figure 2(a), reference environments ER1 and ER2 and a target environment ET are set. For example, the reference environments ER1 and ER2 and the target environment ET may be physically existing environments or may be spatial models that are three-dimensional data reproduced using computer graphics (CG). Alternatively, the environmental parameters of the reference environments ER1 and ER2 may be perturbed within a predetermined range to set environmental parameters for the target environment ET, and a spatial model of the target environment ET may be generated based on the set environmental parameters. Alternatively, an existing real environment or an existing spatial model having environmental parameters whose values ​​are within a predetermined range from the environmental parameters of the reference environments ER1 and ER2 may be used as the target environment ET.

[0028] Next, in Figure 2(b), whether the reference environment ER1 and the target environment ET are similar is determined based on the analogy between the results of a process in which events occurring in the reference environment ER1 are recognized by a predetermined recognition process and the results of a process in which events occurring in the target environment ET are recognized by a recognition process. Then, a similar environment set G1 is formed, which is a set of target environments ET that are similar to the reference environment ER1. Similarly, a similar environment set G2 is formed, which is a set of target environments ET that are similar to the reference environment ER2.

[0029] 2(c), the range of the environmental parameters of the similar environment set G1 (environment range) is calculated based on the environmental parameters of the target environment ET belonging to the similar environment set G1. Similarly, the environment range of the similar environment set G2 is calculated. Then, it is determined whether the environmental parameters of the real environment EA are values ​​within the environmental range of the similar environment set G1 or the similar environment set G2. If the environmental parameters of the real environment EA are values ​​within the environmental range of the similar environment set G1, the processing parameters to be used in the recognition processing in the real environment EA are adjusted using the adjustment parameters that have already been adjusted as processing parameters to be used in the recognition processing in the reference environment ER1. For example, the adjustment parameters that have been adjusted in the reference environment ER1 are set as processing parameters to be used in the recognition processing in the real environment EA.

[0030] Next, the data processing device 20 according to the embodiment will be described. Fig. 3 is a schematic diagram showing an example of the hardware configuration of the data processing device 20 according to the embodiment. The data processing device 20 includes an input unit 21, a communication unit 22, a storage unit 23, a control unit 24, and an output unit 25. Of these, the storage unit 23 and the control unit 24 can be realized by a so-called computer, and the input unit 21, the communication unit 22, and the output unit 25 can be realized as peripheral devices of the computer.

[0031] The input unit 21 includes a user interface such as a keyboard, a mouse, etc. that is operated by a user to input data, etc. The input unit 21 is connected to the control unit 24, converts user operations into operation signals, and outputs the operation signals to the control unit 24. The input unit 21 may also include a DVD drive and a USB interface. The input unit 21 inputs data from these storage media as a file to the control unit 24. The input unit 21 may acquire information about the above-mentioned real environment 2, reference environment, and target environment, and store it in the storage unit 23.

[0032] The communication unit 22 includes a network interface or the like that transmits and receives data between the data processing device 20 and an external device (recognition device 10) via wired or wireless communication. The storage unit 23 is a memory device such as a ROM or RAM, and stores various programs and various data. The control unit 24 is a controller that determines processing parameters used in the recognition process in the recognition device 10, and is configured with an arithmetic unit such as a CPU, a DSP, an MCU, or a GPU.

[0033] The control unit 24 is connected to the storage unit 23, and operates as various processing units by reading and executing computer programs from the storage unit 23, and stores and reads various data in the storage unit 23. The functions of the data processing device 20 described below are realized by the control unit 24 executing the computer programs stored in the storage unit 23. The control unit 24 is also connected to the input unit 21 to accept operation inputs to the data processing device 20 by the user.

[0034] The control unit 24 is connected to the output unit 25 and outputs the processing results of the control unit 24 to the output unit 25. The control unit 24 is also connected to the communication unit 22, and transmits and receives various data to and from external devices via the communication unit 22. For example, the control unit 24 may transmit processing parameters used in the recognition process by the control unit 15 of the data processing device 10 to the recognition device 10 via the communication unit 22.

[0035] 4 is a block diagram showing an example of the functional configuration of the data processing device 20 according to the embodiment. In addition to the storage unit 23, the data processing device 20 includes a target environment acquisition unit 30, a reference environment acquisition unit 31, a recognition processing unit 32, a comparison unit 33, a determination unit 34, an environment range calculation unit 35, a base parameter setting unit 36, a real environment acquisition unit 37, and a processing parameter determination unit 38.

[0036] The storage unit 23 stores environmental parameters indicating a reference environment (hereinafter sometimes referred to as "reference environment parameters") and processing parameters (hereinafter sometimes referred to as "reference processing parameters") adjusted to achieve desired recognition performance in recognition processing in the reference environment. The reference processing parameters may be, for example, processing parameters optimized for recognition processing in the reference environment.

[0037] Furthermore, if the reference environment is a physically existing environment, the storage unit 23 may store image data of the reference environment actually captured by a camera as information about the reference environment. If the reference environment is an existing space model, the storage unit 23 may store parameters of the existing space model of the reference environment as information about the reference environment.

[0038] The storage unit 23 also stores environmental parameters that indicate the target environment (hereinafter, sometimes referred to as "target environment parameters"). If the target environment is a physically existing environment, the storage unit 23 may store image data of the target environment that is actually captured by a camera as information about the target environment. If there is an existing space model that can be used as the target environment, the storage unit 23 may store parameters of the existing space model of the target environment as information about the target environment. The reference environmental parameter and the target environmental parameter are examples of the "third environmental parameter" and the "first environmental parameter" described in the claims.

[0039] The target environment acquisition unit 30 acquires information about the target environment. When the target environment is a physically existing environment, for example, image data of the target environment actually captured by a camera for use in recognition processing in the target environment may be read from the storage unit 23 as information about the target environment. When an existing spatial model that can be used as the target environment exists, parameters of the existing spatial model of the target environment may be read from the storage unit 23, and a CG image may be generated in which an object to be recognized (e.g., a person, a vehicle, etc.) appears in the target environment of the spatial model.

[0040] The target environment acquisition unit 30 may generate a new spatial model of the target environment based on the reference environment parameters. For example, the target environment acquisition unit 30 may perturb the reference environment parameters to generate new target environment parameters. The target environment acquisition unit 30 may generate a spatial model of the target environment based on the new target environment parameters. The target environment acquisition unit 30 is an example of the "environment parameter generation unit" and "target environment generation unit" described in the claims. When a spatial model of a new target environment is generated, the target environment acquisition unit 30 stores the target environment parameters of this target environment in the storage unit 23.

[0041] The reference environment acquisition unit 31 acquires information about the reference environment. When the reference environment is a physically existing environment, for example, image data of the reference environment actually captured by a camera to be used for recognition processing in the reference environment may be read from the storage unit 23 as information about the reference environment. When the reference environment is an existing space model, parameters of the existing space model of the reference environment may be read from the storage unit 23, and a CG image may be generated in which the object to be recognized appears in the reference environment of the space model.

[0042] The recognition processing unit 32 recognizes an event occurring in the reference environment by a predetermined recognition process to obtain a first processing result. For example, the predetermined recognition process may be a recognition process using reference processing parameters adjusted to achieve desired recognition performance in the recognition process in the reference environment. When the reference environment is a physically existing environment, for example, the recognition processing unit 32 may perform a predetermined recognition process using image data within the reference environment acquired by the reference environment acquisition unit 31 as input data as information about the reference environment.

[0043] On the other hand, when the reference environment is a spatial model, for example, the recognition processing unit 32 may perform a predetermined recognition process using a CG image of the reference environment generated by the reference environment acquisition unit 31 as input data. The recognition processing unit 32 obtains a second processing result obtained by recognizing an event occurring in the target environment through a recognition process. The recognition process in the target environment is the same as the recognition process in the reference environment described above. The processing parameters used in the recognition process in the target environment are the same as the reference processing parameters used in the reference environment.

[0044] The comparison unit 33 compares the first processing result in the reference environment with the second processing result in the target environment, and calculates the similarity between the first processing result and the second processing result. For example, the comparison unit 33 may calculate the similarity representing the degree to which the trends of the output results of the recognition processing by the recognition processing unit 32 are similar. The comparison unit 33 may calculate a similarity that indicates how similar the evaluation indices are indicative of the recognition performance by the recognition processing unit 32. For example, when the recognition processing by the recognition processing unit 32 is a person tracking process, the comparison unit 33 may calculate a similarity that indicates how similar the evaluation indices (IDF1, MOTA, etc.) indicative of the tracking performance are between the reference environment and the target environment.

[0045] For example, if the recognition processing by the recognition processing unit 32 is an object detection processing or a person detection processing, the comparison unit 33 may calculate a similarity that represents how similar the evaluation index (mAP, TIDE) indicating the detection performance is between the reference environment and the target environment. Furthermore, for example, if a given recognition process is a combination of multiple recognition logics (including, for example, a learning model or rule-based artificial intelligence), the similarity of the overall processing results of these multiple recognition logics may be calculated, or the similarity of the processing results of some of the recognition logics may be calculated.

[0046] For example, the comparison unit 33 may calculate a similarity that indicates the degree of similarity in the tendency of failure in the recognition process as the similarity in the tendency of the output results of the recognition process by the recognition processing unit 32. For example, the comparison unit 33 may calculate a similarity that represents the degree of similarity between the tendency of locations where false detections occur in the real physical space of the reference environment and the target environment, the tendency of locations in an image where false detections occur, and the tendency of situations where false detections occur.

[0047] Furthermore, for example, the comparison unit 33 may calculate not only the similarity of the final output result of the recognition processing by the recognition processing unit 32, but also the similarity that indicates the degree of similarity of the tendency of the intermediate results of the recognition processing. For example, if the recognition processing by the recognition processing unit 32 is a tracking process that tracks the same object in input images taken at different times, the similarity in the tendency of the detection areas (e.g., bounding boxes) that detected the object on the input images taken at different times to surround the object (e.g., the positions where erroneous recognition occurs, such as the swap described below, the tendency of the situations in which erroneous recognition occurs, the tendency of the amount of deviation from the correct position, the tendency of the size of the surrounding area, etc.) may be calculated as the similarity in the tendency of the intermediate results.

[0048] 5(a) and 5(b) are schematic diagrams illustrating an example of an intermediate result of the person tracking process. Here, it is assumed that the trajectories actually taken by persons P1 and P2 in the input image at times t1, t2, t3, and t4 are the trajectories T1 and T2 shown in FIG. 5(a). In FIG. 5(b), a solid-line rectangle B1 indicates a detection area in which person P1 is detected in the input image, and a dashed-line rectangle B2 indicates a detection area in which person P2 is detected.

[0049] At time t3, the position of detection area B1 deviates from the position of person P1 in the image, and the position of detection area B2 deviates from the position of person P2 in the image. As a result, at the subsequent time t4, a misrecognition occurs in which the detection area B1 for person P1 and the detection area B2 for person P2 are swapped. In this way, the tendency of the detection area to surround the detection target in the image affects how the object tracking process fails. Therefore, by calculating the similarity in the tendency of the detection area to surround the target, the similarity in the tendency of the object tracking process to fail can be estimated.

[0050] Furthermore, the comparison unit 33 may calculate the similarity of the tendency of the detection scores calculated in the recognition process for similar detection targets as the similarity of the tendency of the intermediate results. The comparison unit 33 may calculate a plurality of different similarities based on a plurality of different evaluation methods. For example, the comparison unit 33 may calculate a combination of a plurality of similarities including a similarity of an evaluation index indicating the recognition performance by the recognition processing unit 32, a similarity of a tendency for the recognition processing to fail, and a similarity of an intermediate result of the recognition processing.

[0051] 4, the determination unit 34 determines whether the reference environment and the target environment are similar based on the similarity calculated by the comparison unit 33. For example, if the similarity is equal to or greater than a predetermined threshold, it may be determined that the reference environment and the target environment are similar. When the comparison unit 33 calculates a combination including a plurality of different similarities, it may determine whether the reference environment and the target environment are similar based on the combination of these plurality of similarities. For example, if all similarities of a combination indicate that the first processing result in the reference environment and the second processing result in the target environment are similar, the reference environment and the target environment may be determined to be similar, and if any similarity indicates that the first processing result and the second processing result are dissimilar, the reference environment and the target environment may be determined to be dissimilar.

[0052] 6(a) to 6(e) are schematic diagrams of the similarity judgment between the reference environment and the target environment by the judgment unit 34. Fig. 6(a) shows the evaluation index of the recognition performance of the first processing result in the reference environment ER and the evaluation index of the recognition performance of the second processing result in the target environments ET1 to ET9. FIG. 6(b) shows the positions where the object to be detected was erroneously detected in the recognition process in the reference environment ER, and FIGS. 6(c) to 6(e) show the positions where erroneous detection occurred in the recognition process in the target environments ET3 to ET5.

[0053] 6(b) to 6(e) show the actual physical spaces of the reference environment ER and the target environments ET3 to ET5 as viewed from above, where reference symbols WR, W3, W4, and W5 indicate partitions installed in the reference environment ER and the target environments ET3 to ET5, respectively, reference symbols SR, S3, S4, and S5 indicate architectural structures (e.g., columns) in the reference environment ER and the target environments ET3 to ET5, respectively, and reference symbols FR, F3, F4, and F5 indicate fixtures installed in the reference environment ER and the target environments ET3 to ET5. Circles indicate the locations where false detections occurred.

[0054] Note that the reference environment ER and target environments ET3 to ET5 are separate environments. Therefore, the partitions WR, W3, W4, and W5 are separate partitions, the structures SR, S3, S4, and S5 are separate structures, and the fixtures FR, F3, F4, and F5 are separate fixtures. However, when forming a set of similar environments for a reference environment, a target environment having target environment parameters similar to the reference environment's reference environment parameters is compared with the reference environment, so the partitions WR, W3, W4, and W5, the structures SR, S3, S4, and S5, and the fixtures FR, F3, F4, and F5 have similar layouts.

[0055] See Figure 6(a). As indicated by dashed line 40, the evaluation index of recognition performance in target environments ET3 to ET5 is similar to the evaluation index of recognition performance in reference environment ER. However, as shown in Figures 6(b) to 6(e), while the locations where erroneous detections occur in target environments ET3 and ET4 (i.e., the tendency for recognition processing to fail) are similar to the reference environment, in target environment ET5, erroneous detections occur in locations 41 that are different from the reference environment. Therefore, the determining unit 34 may determine that the target environments ET3 and ET4 among the target environments ET1 to ET9 are similar to the reference environment ER, and may determine that the target environment ET5 is not similar to the reference environment ER.

[0056] See Figure 4. The determination unit 34 determines the similarity between each of a plurality of target environments and the reference environment, and classifies the target environments that are similar to the reference environment, thereby forming a similar environment set, which is a collection of these target environments. The determination unit 34 stores information about the formed similar environment set in the storage unit 23. For example, the determination unit 34 stores in the storage unit 23 the identifiers of the target environments belonging to the similar environment set and the identifiers of the reference environments to which these target environments are similar.

[0057] By repeating the above process, the target environment acquisition unit 30, reference environment acquisition unit 31, recognition processing unit 32, comparison unit 33 and judgment unit 34 form a similar environment set, which is a set of similar environments for each of multiple different reference environments. For example, the reference environment acquisition unit 31 may form a set of similar environments similar to the reference environment ER in the analogy determination described in Fig. 6(a), and then set the target environment ET7 determined to be dissimilar to the reference environment ER as a new reference environment. The recognition processing unit 32, the comparison unit 33, and the determination unit 34 may form a set of similar environments for the newly set reference environment ET7. However, when attempting to determine the similarity between the new reference environment ET7 and other target environments (e.g., target environments ET1, ET2, ET5, ET6, ET8, ET9) using the second processing results of the recognition processing performed in the target environment when forming a set of similar environments for the previous reference environment ER, the recognition performance in the reference environment ET7 is low, resulting in many recognition errors and making it difficult to understand the trend of the second processing results.

[0058] Therefore, the processing parameters in the recognition processing unit 32 may be adjusted so that the recognition processing in the reference environment ET7 achieves desired recognition performance. For example, the processing parameters may be adjusted so as to be optimized for the recognition processing in the reference environment ET7. The recognition processing unit 32 may store the processing parameters adjusted for the reference environment ET7 in the storage unit 23 as reference processing parameters for the reference environment ET7.

[0059] See FIG. 4. The environment range calculation unit 35 calculates the environment range of the environment parameters for each of the similar environment sets formed by the determination unit 34. For example, the environment range calculation unit 35 may calculate the environment range of the similar environment set based on the target environment parameters of the target environment belonging to the similar environment set. For example, the environment range calculation unit 35 may calculate the distribution range of the target environment parameters of the target environment belonging to the similar environment set as the environment range. The environment range calculation unit 35 stores the calculated environment range in the memory unit 23.

[0060] The base parameter setting unit 36 ​​sets, for each similar environment set, processing parameters suitable for recognition processing in an environment having environmental parameters whose values ​​are within the environmental range of the similar environment set as “base parameters” and stores them in the storage unit 23. For example, the base parameter setting unit 36 ​​may set, as the base parameters, the reference processing parameters of a reference environment that is similar to the target environment that belongs to the similar environment set.

[0061] The base parameter setting unit 36 ​​may set the base parameters by dividing them into common parameters having values ​​common to environments having environmental parameters whose values ​​are within the environmental range of the similar environment set, and specialized parameters that are set to different values ​​depending on each of these environments. For example, the base parameter setting unit 36 ​​may distinguish between base parameters to be set as common parameters and base parameters to be set as specialized parameters based on an analogy between a first processing result in a reference environment and a second processing result in a target environment.

[0062] For example, processing parameters are adjusted in advance for each target environment so that the recognition processing in each target environment belonging to the similar environment set achieves the desired recognition performance. For example, the processing parameters may be adjusted so that the recognition processing in the target environment is optimized. In the following description, processing parameters adjusted for each target environment belonging to the similar environment set in this way are referred to as "specific processing parameters." The specific processing parameters may be prepared for all target environments belonging to the similar environment set, or may be prepared only for a portion of multiple target environments.

[0063] The base parameter setting unit 36 ​​may extract target environments that have a relatively low degree of similarity from among the target environments that belong to the similar environment set, based on the degree of similarity between the first processing result in the reference environment and the second processing result in the target environment. The base parameter setting unit 36 ​​calculates an environmental parameter difference, which is the deviation between the extracted target environmental parameter and the reference environmental parameter of the reference environment, and also calculates a processing parameter difference, which is the difference between the extracted unique processing parameter of the target environment and the reference processing parameter of the reference environment.

[0064] For example, the base parameter setting unit 36 ​​may distinguish between common parameters and specialized parameters based on the processing parameter difference. For example, the base parameter setting unit 36 ​​may set processing parameters whose processing parameter difference is equal to or greater than a threshold as specialized parameters, and set processing parameters whose processing parameter difference is less than the threshold as common parameters. The base parameter setting unit 36 ​​determines which specialized parameters should be adjusted in response to the fluctuations in the environmental parameters based on the environmental parameter differences and the processing parameter differences. It also calculates the adjustment amounts of the specialized parameters in response to the fluctuations in the environmental parameters based on the environmental parameter differences and the processing parameter differences. The base parameter setting unit 36 ​​stores information about the specialized parameters to be adjusted in response to the fluctuations in the environmental parameters and the adjustment amounts in the storage unit 23 as base parameter adjustment information.

[0065] As another method for distinguishing between common parameters and specialized parameters, correspondences that determine, for each environmental parameter, the extent to which a processing parameter should be adjusted in response to a variation in that environmental parameter may be investigated in advance and stored in the storage unit 23. In this case, the base parameter setting unit 36 ​​may set, as specialized parameters, processing parameters that are associated with environmental parameters whose environmental parameter differences are equal to or greater than a threshold, and set other processing parameters as common parameters.

[0066] The real environment acquisition unit 37 (the "acquisition unit" in the present invention) acquires information about the real environment 2 to which the recognition system 1 is applied. For example, the real environment acquisition unit 37 may acquire environmental parameters of the real environment 2 as information about the real environment 2. In the following description, the environmental parameters of the real environment 2 may be referred to as "real environment parameters." The real environment 2 is an example of "another environment" as defined in the claims, and the real environment parameters are an example of "second environmental parameters."

[0067] The processing parameter determination unit 38 determines processing parameters to be used in the recognition process for recognizing events occurring in the real environment 2, based on the real environment parameters, the environmental range of the similar environment set, and the base parameters. In the following description, the processing parameters to be used in the recognition process for recognizing events occurring in the real environment 2 may be simply referred to as "processing parameters of the real environment 2." For example, the processing parameter determination unit 38 may compare the real environment parameters with each of the environmental ranges of a plurality of similar environment sets stored in the storage unit 23, and identify a similar environment set for which an environmental range that includes the value of the real environment parameter is calculated. The processing parameter determination unit 38 may set processing parameters for the real environment 2 based on base parameters of the identified similar environment set. For example, the base parameters of the identified similar environment set may be set as processing parameters for the real environment 2.

[0068] Furthermore, the processing parameter determination unit 38 may adjust specialized parameters of the base parameters set as processing parameters in accordance with the real environment parameters and set them as processing parameters for the real environment 2. For example, the processing parameter determination section 38 may adjust the specialized parameters according to the difference between the reference environment parameters and the actual environment parameters.

[0069] 7 is a schematic diagram illustrating an example of a method for adjusting specialized parameters of base parameters. The circular plots indicate the positions of the reference environment parameters of the reference environment ER in a parameter space with the axes of environmental parameters 1 and 2, and the triangular plots indicate the positions of the real environment parameters of the real environment EA. For ease of explanation, FIG. 7 shows an example of a two-dimensional parameter space with the axes of two types of environmental parameters, but the type of parameters included in the environmental parameters may be one or three or more.

[0070] As shown in the figure, the difference in environmental parameter 1 between the reference environment ER1 and the actual environment EA is Δ1, and the difference in environmental parameter 2 is Δ2. For example, the processing parameter determination unit 38 may identify specialized parameters to be adjusted in response to fluctuations in environmental parameter 1 and environmental parameter 2, based on base parameter adjustment information stored in the storage unit 23. The processing parameter determination unit 38 may calculate the adjustment amount of the identified specialized parameters, based on the differences Δ1 and Δ2 and the base parameter adjustment information.

[0071] (operation) FIG. 8 is a flowchart of an example of a data processing method according to the embodiment. In step S1, a reference environment is determined, and the reference environment parameters of the determined reference environment, the reference processing parameters, and the image data and space model within the reference environment are stored in the storage unit . In step S2, the target environment acquisition unit 30 perturbs the reference environment parameters of the reference environment to generate new target environment parameters of the target environment, and generates a spatial model of the target environment. The target environment acquisition unit 30 may acquire an existing spatial model or an existing environment as the target environment.

[0072] In step S3, the recognition processing unit 32 recognizes events occurring in the reference environment and the target environment by a predetermined recognition process to obtain a first processing result and a second processing result. The comparison unit 33 calculates the similarity between the first processing result and the second processing result. In step S4, the determination unit 34 classifies the target environments based on the similarity calculated by the comparison unit 33, and forms a set of similar environments.

[0073] In step S5, the environment range calculation unit 35 calculates the environment range of the environment parameters for each of the similar environment sets formed by the determination unit . In step S6, the base parameter setting unit 36 ​​sets base parameters for each set of similar environments.

[0074] In step S7, the real environment acquisition unit 37 acquires real environment parameters of the real environment 2. The processing parameter determination unit 38 determines a similar environment set in which an environment range that includes the value of the real environment parameter is calculated. In step S8, the processing parameter determination unit 38 determines processing parameters for the actual environment 2 based on the determined base parameters of the set of similar environments, after which the processing ends.

[0075] (First Modification) In the above description, an embodiment has been exemplified in which the processing parameter determination unit 38 determines, based on the environmental range of the similar environment set, which of the base parameters of the similar environment set the processing parameters of the actual environment 2 should be based on. Alternatively, the recognition processing unit 32 may recognize an event occurring in the real environment 2 by a predetermined recognition processing and output the processing result. The comparison unit 33 and the determination unit 34 may determine whether the real environment 2 and the reference environment are similar based on an analogy between the processing result of the recognition processing in the real environment 2 and the first processing result of the recognition processing in the reference environment. The processing parameter determination unit 38 may set processing parameters for the real environment 2 based on base parameters set using standard processing parameters of a reference environment similar to the real environment 2. The actual environment 2 in the first modified example is an example of the "target environment" in the claims.

[0076] (Second Modification) In the above description, an embodiment has been exemplified in which the processing parameter determination unit 38 determines, based on the environmental range of the similar environment set, which of the base parameters of the similar environment set the processing parameters of the actual environment 2 should be based on. Alternatively, the processing parameter determination unit 38 may determine the similarity between the real environment parameters of the real environment 2 and the reference environment parameters of the reference environment, and identify the reference environment parameters similar to the real environment parameters as the reference environment similar to the real environment 2. The processing parameter determination unit 38 may set the processing parameters of the real environment 2 based on the reference processing parameters of the reference environment similar to the real environment 2.

[0077] (Effects of the embodiment) (1) The data processing device 20 includes a comparison unit 33 that determines the similarity between a first processing result obtained by recognizing an event occurring in a reference environment through a predetermined recognition process and a second processing result obtained by recognizing an event occurring in a target environment through a recognition process, and a determination unit 34 that determines whether the reference environment and the target environment are similar based on the similarity between the first processing result and the second processing result. This allows for determining the similarity between environments that are the subject of a given recognition process, and for determining whether these environments are similar to each other for the given recognition process.

[0078] (2) The determination unit 34 may form a set of target environments similar to the reference environment by determining the similarity between each of the target environments and the reference environment, thereby estimating the range of environments similar to the reference environment. (3) The data processing device may include a storage unit 23 that stores first environmental parameters, which are environmental parameters indicating a target environment, and an environmental range calculation unit 35 that calculates an environmental range, which is a range of environmental parameters, based on the first environmental parameters of the target environment belonging to the set formed by the determination unit 34. This makes it possible to calculate the range of environmental parameters of an environment similar to the reference environment.

[0079] (4) The storage unit 23 may store base parameters (e.g., reference processing parameters of the reference environment) that are processing parameters used in the recognition process to recognize an event occurring in the reference environment. The data processing device 20 may include a processing parameter determination unit 38 that determines the base parameters as processing parameters to be used in the recognition process to recognize an event occurring in the other environment when a second environmental parameter, which is an environmental parameter indicating an environment other than the target environment, has a value within the environmental range. This allows the processing parameters that have been adjusted for the reference environment to be used to adjust the parameters of the recognition process in other environments.

[0080] (5) The processing parameter determination unit 38 may determine processing parameters to be used in the recognition process in the other environment by correcting the base parameters based on the difference between the second environmental parameters representing the other environment and the third environmental parameters representing the reference environment. This allows the processing parameters to be customized according to the characteristics unique to the other environment.

[0081] (6) The data processing device 20 may include a storage unit 23 that stores base parameters, which are processing parameters used in a recognition process that recognizes an event occurring in a reference environment, and a processing parameter determination unit 38 that determines the base parameters as processing parameters to be used in a recognition process that recognizes an event occurring in a target environment that has been determined to be similar to the reference environment by the determination unit 34. This allows the processing parameters in the target environment to be set by utilizing the processing parameters in the reference environment that has been determined to be similar to the target environment for the recognition process.

[0082] (7) The data processing device 20 may include an environmental parameter generation unit that generates first environmental parameters, which are environmental parameters representing a target environment, based on third environmental parameters, which are environmental parameters representing a reference environment, and a target environment generation unit that generates a spatial model of the target environment based on the first environmental parameters generated by the environmental parameter generation unit. The comparison unit 33 may acquire the second processing result by performing a recognition process using the spatial model generated by the target environment generation unit. This allows the target environment to be virtually generated, thereby reducing the cost of building the target environment.

[0083] (8) The determination unit 34 may determine whether the reference environment and the target environment are similar based on the similarity between the first processing result and the second processing result. The similarity may represent the degree of similarity between the output result or the intermediate result of the recognition process. This makes it possible to determine whether the reference environment and the target environment are similar for recognition processing.

[0084] (9) The environmental parameters may include at least one of a parameter representing the installation height of a camera capturing an image of a target of recognition processing, a parameter representing the material of an object captured by the camera, a parameter representing the monitoring range of the target of recognition processing, a parameter representing the number of targets of recognition processing, and a parameter representing the arrangement of furniture installed in the environment of the target of recognition processing. By using such environmental parameters, it is possible to determine whether the environments are similar to each other.

[0085] (10) The data processing device 20 may include a memory unit 23 that stores a third environmental parameter, which is an environmental parameter indicating a reference environment, and a base parameter, which is a processing parameter used in a recognition process to recognize an event occurring in the reference environment; a judgment unit 34 that determines whether the other environment is similar to the reference environment based on an analogy between the second environmental parameter, which is an environmental parameter indicating an environment other than the reference environment, and the third environmental parameter; and a processing parameter determination unit 38 that, when it is determined that the other environment is similar to the reference environment, determines the base parameter as a processing parameter to be used in a recognition process to recognize an event occurring in the other environment. This allows the processing parameters that have been adjusted for the reference environment to be used to adjust the parameters of the recognition process in other environments. [Explanation of symbols]

[0086] 1...recognition system, 2...real environment, 3...sensor (camera), 10...recognition device, 12, 21...input unit, 13, 22...communication unit, 14, 23...storage unit, 15, 24...control unit, 16, 24...output unit, 20...data processing device, 30...target environment acquisition unit, 31...reference environment acquisition unit, 32...recognition processing unit, 33...comparison unit, 34...determination unit, 35...environment range calculation unit, 36...base parameter setting unit, 37...real environment acquisition unit, 38...processing parameter determination unit

Claims

1. a comparison unit that determines the similarity between a first processing result obtained by recognizing an event occurring in a reference environment through a predetermined recognition process and a second processing result obtained by recognizing an event occurring in a target environment through the recognition process; a determination unit that determines whether the reference environment and the target environment are similar based on an analogy between the first processing result and the second processing result; A data processing device comprising:

2. The data processing device according to claim 1 , wherein the determining unit determines an analogy between each of the plurality of target environments and the reference environment, thereby forming a set of the target environments that are similar to the reference environment.

3. a storage unit that stores a first environmental parameter that is an environmental parameter indicating the target environment; an environment range calculation unit that calculates an environment range, which is a range of environment parameters, based on the first environment parameters of the target environments that belong to the set formed by the determination unit; 3. The data processing apparatus according to claim 2, further comprising:

4. the storage unit stores base parameters, which are processing parameters used in a recognition process for recognizing an event occurring in the reference environment; The data processing device includes: an acquisition unit that acquires a second environmental parameter that is an environmental parameter indicating an environment other than the target environment that belongs to the set; a processing parameter determination unit that, when the acquired second environmental parameter is a value within the environmental range, sets the base parameter as a processing parameter used in a recognition process for recognizing an event occurring in the other environment; 4. The data processing apparatus according to claim 3, further comprising:

5. 5. The data processing device according to claim 4, wherein the processing parameter determination unit determines the processing parameters to be used in the recognition process in the other environment by correcting the base parameters based on a difference between a second environmental parameter indicating the other environment and a third environmental parameter which is an environmental parameter indicating the reference environment.

6. a storage unit that stores base parameters, which are processing parameters used in a recognition process that recognizes an event occurring in the reference environment; a processing parameter determination unit that determines the base parameters as processing parameters to be used in a recognition process for recognizing an event occurring in the target environment determined by the determination unit to be similar to the reference environment; 2. The data processing device according to claim 1, further comprising:

7. an environmental parameter generation unit that generates a first environmental parameter that is an environmental parameter representing the target environment based on a third environmental parameter that is an environmental parameter representing the reference environment; a target environment generation unit that generates a spatial model of the target environment based on the first environmental parameters generated by the environmental parameter generation unit; Equipped with The data processing device according to any one of claims 1 to 5, characterized in that the comparison unit acquires the second processing result by performing the recognition process using a spatial model generated by the target environment generation unit.

8. the determination unit determines whether the reference environment and the target environment are similar based on a degree of similarity between the first processing result and the second processing result; 7. The data processing device according to claim 1, wherein the degree of similarity represents a degree of similarity in the tendency of output results or intermediate results of the recognition processing.

9. 4. The data processing device according to claim 3, wherein the environmental parameters include at least one of a parameter representing the installation height of a camera that photographs the target of the recognition processing, a parameter representing the monitoring range of the target of the recognition processing, a parameter representing the number of targets of the recognition processing, a parameter representing the arrangement of objects installed in the environment of the target of the recognition processing, and a parameter representing the lighting conditions in the environment of the target of the recognition processing.

10. a storage unit that stores third environmental parameters that are environmental parameters indicating a reference environment and base parameters that are processing parameters used in a recognition process that recognizes an event that occurs in the reference environment; a determination unit that determines whether the other environment is similar to the reference environment based on a comparison between a second environmental parameter, which is an environmental parameter indicating the other environment, and the third environmental parameter; a processing parameter determination unit that, when it is determined that the other environment is similar to the reference environment, determines the base parameters as processing parameters to be used in a recognition process for recognizing an event occurring in the other environment; A data processing device comprising:

11. determining an analogy between a first processing result obtained by recognizing an event occurring in a reference environment through a predetermined recognition process and a second processing result obtained by recognizing an event occurring in a target environment through the same recognition process; determining whether the reference environment and the target environment are similar based on the similarity between the first processing result and the second processing result; A data processing method comprising:

12. storing in advance in a storage device third environmental parameters which are environmental parameters indicating a reference environment and base parameters which are processing parameters used in a recognition process for recognizing an event occurring in the reference environment; determining whether the other environment is similar to the reference environment based on a comparison between a second environmental parameter, which is an environmental parameter indicating the other environment, and the third environmental parameter; If it is determined that the other environment is similar to the reference environment, the base parameters are determined as processing parameters to be used in a recognition process for recognizing an event occurring in the other environment. A data processing method comprising:

Images