Entry inference system
The entry estimation system enhances user entry prediction accuracy by integrating environmental and temporal data into an AI model, addressing the limitations of existing systems.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- SUMITOMO ELECTRIC INDUSTRIES LTD
- Filing Date
- 2025-11-05
- Publication Date
- 2026-07-23
AI Technical Summary
Existing systems struggle to accurately estimate user entry into specific areas, such as crosswalks, due to insufficient consideration of environmental and temporal conditions.
An entry estimation system that utilizes a detection unit to gather data on user presence and behavior, combined with condition data like time, weather, and season, and inputs this information into an AI model to enhance estimation accuracy.
Improves the accuracy of entry estimation by considering multiple factors, allowing for precise control of traffic signals and vehicle notifications based on user intentions.
Smart Images
Figure JP2025038786_23072026_PF_FP_ABST
Abstract
Description
Entry Estimation System
[0001] The present disclosure relates to an entry estimation system. This application claims priority based on Japanese Application No. 2025-005348 filed on January 15, 2025, and incorporates all the descriptions set forth in the above Japanese application.
[0002] Patent Document 1 discloses a technique for identifying a person waiting to cross a crosswalk by analyzing the line of sight of a person in a waiting area waiting for a signal at a crosswalk and controlling a traffic signal related to the crosswalk.
[0003] Japanese Patent Application Laid-Open No. 2017-208141
[0004] An entry estimation system according to an embodiment includes a detection unit that detects a user in a first area and outputs a detection result, and a control unit that estimates whether the user enters from the first area into a second area. The control unit includes an acquisition unit that acquires condition data including the detection result and one or more conditions related to the detection of the user by the detection unit, a derivation unit that derives feature amount data including one or more input feature amounts based on the detection result, an estimation unit, and an output unit. The estimation unit obtains an estimation result by inputting the feature amount data derived by the derivation unit and the condition data acquired by the acquisition unit into an AI model that outputs an estimation result as to whether the user enters from the first area into the second area according to a combination of the feature amount data and the condition data. The output unit outputs the estimation result by the estimation unit.
[0005] FIG. 1 is a diagram for explaining an overview of the entry estimation system according to this embodiment. FIG. 2 is a block diagram illustrating the hardware configuration of the control unit. FIG. 3 is a functional block diagram of the control unit. FIG. 4 is a diagram for explaining an overview of entry estimation. FIG. 5 is a diagram for explaining details of entry estimation. FIG. 6 is a diagram for explaining an application scenario of the entry estimation system. FIG. 7 is a flowchart for explaining an entry estimation process. FIG. 8 is a diagram for explaining entry estimation according to a comparative example. FIG. 9 is a diagram for explaining entry estimation according to a comparative example.
[0006] Here, in the conventional method, there were cases where it was not possible to estimate with sufficient accuracy whether a user enters a specific area such as a crosswalk.
[0007] This disclosure is made in light of the above circumstances and aims to improve the accuracy of estimating entry into a specific area.
[0008] According to this disclosure, it is possible to improve the accuracy of estimating entry into a specific area.
[0009] First, the embodiments of this disclosure will be listed and explained.
[0010] [1] An entry estimation system according to one embodiment includes a detection unit that detects a user in a first area and outputs a detection result, and a control unit that estimates whether or not the user enters a second area from the first area. The control unit includes an acquisition unit that acquires the detection result and condition data including one or more conditions related to the detection of the user by the detection unit, a derivation unit that derives feature data including one or more input features based on the detection result, an estimation unit, and an output unit. The estimation unit obtains an estimation result by inputting the feature data derived by the derivation unit and the condition data acquired by the acquisition unit into an AI model that outputs an estimation result of whether or not the user enters a second area from the first area according to a combination of feature data and condition data. The output unit outputs the estimation result by the estimation unit.
[0011] In this entry estimation system, feature data including one or more input features derived according to the detection result, and condition data acquired by the acquisition unit are input to the AI model, and an estimation result of whether or not the user enters the second area from the first area is output according to the combination of feature data and condition data. In this way, an estimation result corresponding to the combination of feature data and condition data according to the detection result is output. As a result, the entry estimation system can output estimation results with higher accuracy compared to a case where the estimation result is output simply from feature data according to the detection result (e.g., the user's trajectory), because conditions related to user detection (e.g., time of day, weather, etc.) are taken into consideration. For example, if there is a route that a user is likely to take when entering the second area in rainy weather, the entry estimation system can estimate that the user will enter the second area from a combination of feature data (the user's trajectory follows that route) and condition data (it is raining), and can perform entry estimation with high accuracy by taking condition data into consideration. Therefore, the entry estimation system according to this embodiment can improve the accuracy of entry estimation to a specific area.
[0012] [2] In the approach estimation system described in [1] above, the control unit may further include a learning unit that learns an AI model by associating feature data, condition data, and estimation results. By learning the feature data, condition data, and estimation results in this way, the estimation accuracy of the AI model can be appropriately improved.
[0013] [3] In the entry estimation system described in [1] or [2] above, the condition data may include data that includes the time period at the time of detection by the detection unit. This allows the entry estimation system to perform entry estimation with high accuracy by taking into account the characteristics of each time period (for example, a certain route is more likely to be adopted by users entering the second area in the evening or later).
[0014] [4] In the entry estimation system described in any one of the above items [1] to [3], the condition data may include data that includes the month or season conditions at the time of detection by the detection unit. This allows the entry estimation system to perform entry estimation with high accuracy by taking into account seasonal characteristics (for example, a certain area may be shaded by trees and therefore likely to be used as a waiting position for users entering the second area during the hot summer months).
[0015] [5] In the entry estimation system described in any one of the above items [1] to [4], the condition data may include data on the temperature at the time of detection by the detection unit or within a predetermined time from the time of detection. This allows the entry estimation system to perform entry estimation with high accuracy by taking into account the characteristics of the temperature (for example, a certain area may be shaded by trees and therefore more likely to be used as a waiting position for users entering the second area when the temperature is high).
[0016] [6] In the entry estimation system described in any one of the above items [1] to [5], the condition data may include data on the weather conditions at the time of detection by the detection unit or within a predetermined time from the time of detection. This allows the entry estimation system to perform entry estimation with high accuracy by taking into account weather characteristics (for example, a certain area does not get wet in the rain and is therefore likely to be used as a waiting position for users entering the second area when it is raining).
[0017] [7] In the entry estimation system described in any one of [1] to [6] above, the derivation unit may derive at least one of the following as input features based on the detection result: the user's trajectory, the user's type, the orientation of the user's face, environmental data of the area where the user is located, and the distance from the user to the second area. This allows the entry estimation system to appropriately estimate whether or not the user will enter the second area.
[0018] Embodiments of this disclosure will be described with reference to the drawings. This disclosure is not limited to these examples, but is indicated by the claims, and all modifications within the meaning and scope of the claims are intended to be included. In the description of the drawings, identical elements are denoted by the same reference numerals, and redundant descriptions are omitted.
[0019] Figure 1 is a diagram illustrating the overview of the entry estimation system 1 according to this embodiment.
[0020] The entry estimation system 1 estimates whether a user will enter the second area from the first area based on the detection image, which is the result of the camera 20 (detection unit) detecting a user in the first area. In this embodiment, an area including a pedestrian crossing E1 is given as an example of the second area. A sidewalk E2 adjacent to the pedestrian crossing E1 is given as an example of the first area. That is, in this embodiment, the first area and the second area are adjacent to each other. Sidewalk E2 includes an area that user Y passing through immediately before entering the pedestrian crossing E1, and also includes an area where user Y waiting when a vehicle is passing over the pedestrian crossing E1. For example, user Y waits at a traffic light on sidewalk E2, then enters the pedestrian crossing E1, crosses the pedestrian crossing E1, and reaches the sidewalk on the opposite side. Examples of user Y may include not only people walking, but also people moving on vehicles (bicycles, wheelchairs, kick scooters, strollers, etc.). Examples of user Y may also include people walking while using their smartphones (smartphone zombies), people with white canes, etc. User Y may include moving objects other than people (animals, robots, etc.).
[0021] The entry estimation system 1 can divide users Y detected on the sidewalk E2 into users who intend to cross (intend to enter) the pedestrian crossing E1 and users who do not intend to cross (intend to enter). The entry estimation system 1 estimates whether user Y will enter the pedestrian crossing E1 by, for example, estimating the intention to cross of the detected user Y. As shown in Figure 1, if there are multiple pedestrian crossings E1 that can be entered from the sidewalk E2, the entry estimation system 1 may estimate which pedestrian crossing E1 user Y is trying to enter. If the entry estimation system 1 estimates that there is a user Y who will enter the pedestrian crossing E1, it may control the traffic signal 60 (one component of the external device 5) related to the pedestrian crossing E1 to change the color of the signal or extend or shorten the green light time. If the entry estimation system 1 estimates that there is a user Y who will enter the pedestrian crossing E1, it may notify the on-board device of a vehicle C (one component of the external device 5) traveling around the pedestrian crossing E1. The traffic signals controlled by the entry estimation system 1 may be traffic signals 60 located in Japan as described above, or they may be traffic signals located in foreign countries.
[0022] As shown in Figure 1, the entry estimation system 1 comprises a control unit 10 and a camera 20 (detection unit). In the example shown in Figure 1, the control unit 10 and the camera 20 are separate components, but the functions of the control unit 10 may be incorporated into the camera 20 (the control unit 10 and the camera 20 may be integrated into a single unit).
[0023] Camera 20 is installed on the sidewalk E2. More specifically, camera 20 is installed on a pole 50 located on the sidewalk E2. Pole 50 is a signal pole on which a traffic light 60 is installed. Pole 50 may also be a signpost or a lighting pole. Camera 20 may be installed near the top of pole 50.
[0024] Camera 20 is a detection unit that detects a user Y passing through the sidewalk E2 and outputs a detection image as the detection result. Camera 20 is installed directly above the sidewalk E2, and by setting the detection range downward from the perspective of camera 20, it detects user Y on the sidewalk E2. Directly above the sidewalk E2 means vertically above a part of the sidewalk E2. Camera 20 does not have to be installed directly above the sidewalk E2; it may be installed diagonally above the sidewalk E2 as long as it can detect the entire area of the sidewalk E2. The sidewalk E2 is set up in a rectangular shape in plan view of several meters x several meters, for example, it may be set up as an area of 4 meters x 4 meters.
[0025] Camera 20 only needs to have its horizontal and vertical angles set so that it can detect the entire sidewalk E2, and the horizontal and vertical angles may be set to approximately ±20°. Sidewalk E2 may be the entire detection range of camera 20, or only a part of it. In Figure 1, two cameras 20A and 20B are shown as examples of camera 20. Camera 20A has the left sidewalk E2 of the crosswalk E1 in the figure as its detection range D1. Camera 20B has the right sidewalk E2 of the crosswalk E1 in the figure as its detection range D2.
[0026] Camera 20 periodically captures images of the sidewalk E2 at predetermined time intervals and continuously transmits the detection images, which are the detection results, to the control unit 10. Camera 20 transmits the detection results to the control unit 10 via a base station (not shown), or via direct wireless communication or wired communication that does not go through a base station.
[0027] In this embodiment, a camera 20 is used as an example of the detection unit, but it is not limited to this. The detection unit may be a configuration other than a camera 20, as long as it can detect the user Y. The detection unit may also be a detection device such as a laser or lidar. A millimeter-wave radar 80, as shown in Figure 1, may be installed on the pole 50. In the example shown in Figure 1, the millimeter-wave radar 80 has the area of the pedestrian crossing E1 as its detection range D3.
[0028] The control unit 10 is implemented by one or more control computers. Figure 2 is a block diagram illustrating the hardware configuration of the control unit 10. The control unit 10 has the circuit 120 shown in Figure 2. The circuit 120 has one or more processors 121, a memory 122, a storage 123, and an input / output port 124. The storage 123 has a storage medium that can be read by a computer, such as a hard disk. The storage medium stores a program for executing a predetermined intrusion estimation processing procedure. The storage medium may be a removable medium such as a non-volatile semiconductor memory, a magnetic disk, or an optical disk. The memory 122 temporarily stores the program loaded from the storage medium of the storage 123 and the calculation results by the processor 121. The processor 121 implements each of the functional modules described later by executing the above program in cooperation with the memory 122. The input / output port 124 inputs and outputs electrical signals according to commands from the processor 121.
[0029] The hardware configuration of the control unit 10 is not limited to implementing each functional module by program. Each functional module of the control unit 10 may be implemented by a processing circuit. The processing circuit may consist of a processor, an integrated circuit, or other circuit that combines various analog circuits and various digital circuits. The processor may be any processor suitable for computer control, such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), DSP (Digital Signal Processor), FPGA (Field Programmable Gate Array), or ASIC (Application Specific Integrated Circuit).
[0030] The control unit 10 estimates whether user Y enters the crosswalk E1 from the sidewalk E2. Figure 3 is a functional block diagram of the control unit 10. As shown in Figure 3, the control unit 10 includes an acquisition unit 11, a derivation unit 12, an estimation unit 13, an output unit 14, a learning unit 15, and a storage unit 16. The storage unit 16 stores an AI model 100 (details will be described later) that outputs the estimation results. The AI model 100 is, for example, Transformer, Social GAN, or Social LSTM.
[0031] The overview of entry estimation by the control unit 10 will be explained. Figure 4 is a diagram illustrating the overview of entry estimation. As shown in Figure 4, the control unit 10 estimates whether user Y enters the crosswalk E1 from the sidewalk E2 by inputting various input features into the AI model 100 described above. The input features are derived based on the detection results of user Y on the sidewalk E2 by the camera 20. The input features are, for example, user Y's trajectory (time-series x and y coordinates), pedestrian (user Y) type, user Y's face orientation, user Y's posture, environmental data of the area where user Y is located (whether it is a pedestrian area or not, etc.), or the distance from user Y to the exit of the crosswalk E1 (a position on the crosswalk E1 that is continuous with sidewalk E2). The output from the AI model 100 may be the probability that user Y enters the exit of each crosswalk E1.
[0032] Figure 5 is a diagram illustrating the details of the entry estimation. As shown in Figure 5, in the control unit 10, not only the input features described above are directly input to the AI model 100, but a combination of input features and conditional data is input to the AI model 100. This enables the training of the AI model 100 and the output of estimation results using the trained AI model. Conditional data is data that includes one or more conditions related to the detection of user Y by the camera 20. Conditional data includes data such as the time of day when detection is performed by the camera 20, the month or season when detection is performed by the camera 20, the temperature at the time of detection by the camera 20 or within a predetermined time (e.g., within 15 minutes) from detection, and the weather at the time of detection by the camera 20 or within a predetermined time (e.g., within 15 minutes) from detection. In this way, by outputting estimation results corresponding to the combination of input features and conditional data according to the detection result, the estimation results can be output with higher accuracy compared to the case where the estimation result is simply output from input features according to the detection result alone, because conditions related to the detection of user Y (e.g., time of day, weather, etc.) are taken into consideration.
[0033] Figure 6 illustrates the application scenarios of the entry estimation system 1. Figure 6 shows specific examples I to IV as examples of scenarios in which the entry estimation system can be effectively utilized. In specific example I, when children returning home from elementary school after 5 PM tend to take a specific route home and cross a specific crosswalk E1, an AI model 100 is prepared in advance, having learned combinations of detection results in the sidewalk E2 surrounding crosswalk E1 and the time of day. From this combination, it can be estimated that the detected user Y will enter crosswalk E1. In specific example II, when user Y tends to wait in the shaded area of sidewalk E2 around 12 PM on a sunny summer day, an AI model 100 is prepared in advance, having learned combinations of detection results in the shaded area of sidewalk E2 and the season and time of day. From this combination, it can be estimated that the detected user Y will enter crosswalk E1. In specific example III, when it is raining, user Y tends to wait under a tree on sidewalk E2 before crossing pedestrian crossing E1. An AI model 100 is prepared in advance, having learned combinations of detection results in the area of the tree on sidewalk E2 and the weather: rain. From this combination, it can be estimated that the detected user Y will enter pedestrian crossing E1. In specific example IV, when it is raining the day before, puddles tend to form in a specific location on sidewalk E2, and user Y tends to approach pedestrian crossing E1 by taking a detour around the puddle. An AI model 100 is prepared in advance, having learned combinations of detection results along the detour route and the weather: it rained the previous day. From this combination, it can be estimated that the detected user Y will enter pedestrian crossing E1.
[0034] Returning to Figure 3, the functions implemented in the control unit 10 will be described. The acquisition unit 11 acquires detection results from the camera 20 and also from the millimeter-wave radar 80. The following description will omit the specific details of how the detection results from the millimeter-wave radar 80 are used, but they may be used as appropriate for approach estimation as detection results at the pedestrian crossing E1. The acquisition unit 11 may acquire data each time data is transmitted from the camera 20 and the millimeter-wave radar 80, or it may acquire data in batches at predetermined time intervals. Along with the detection results, the acquisition unit 11 acquires condition data including the position or multiple conditions related to the detection of user Y by the camera 20 (and the millimeter-wave radar 80). As described above, the condition data may include data such as the time period conditions at the time of detection by the camera 20, the month or season conditions at the time of detection by the camera 20, the temperature conditions at the time of detection by the camera 20 or within a predetermined time from detection, and the weather conditions at the time of detection by the camera 20 or within a predetermined time from detection.
[0035] The derivation unit 12 derives feature data including one or more input features based on the detection results. Based on the detection results, the derivation unit 12 may derive at least one of the following as input features: the trajectory of user Y, the type of user Y, the orientation of user Y's face, environmental data of the area where user Y is located, and the distance from user Y to the exit of the pedestrian crossing E1. Such derivation may be performed using conventionally known image recognition techniques. The input features derived by the derivation unit 12 may be designed to reduce the burden on subsequent estimation processing, and the images do not necessarily have to be used as they are. By using data other than the trajectory as input features, it becomes possible to estimate users waiting for taxis or to estimate from a distance. By using the above-mentioned environmental data as input features, it becomes possible to consider, for example, whether or not an area is passable for user Y. By using the distance to the pedestrian crossing E1 as input features, the accuracy of entry estimation can be further improved.
[0036] The memory unit 16 stores an AI model 100 that outputs an estimation result of whether or not user Y enters the crosswalk E1 from the sidewalk E2, depending on the combination of feature data and condition data. The AI model 100 is an AI model that has been pre-trained by the learning unit 15 or an external learning configuration.
[0037] The learning unit 15 learns the AI model 100 by associating the feature data derived by the derivation unit 12, the condition data acquired by the acquisition unit 11, and the estimation results (the estimation results obtained by the estimation unit 13 using the AI model 100).
[0038] The estimation unit 13 obtains estimation results by inputting the feature data derived by the derivation unit 12 and the condition data acquired by the acquisition unit 11 into the AI model 100.
[0039] The output unit 14 outputs the estimation result from the estimation unit 13. The output unit 14 may also control the external device 5 by outputting the estimation result to the external device 5. In this case, the external device 5 may be, for example, a traffic signal 60 related to the pedestrian crossing E1, or an on-board device of a vehicle C traveling around the pedestrian crossing E1.
[0040] Figure 7 is a flowchart illustrating the entry estimation process. As shown in Figure 7, first, the acquisition unit 11 acquires the detection result and condition data (step S1).
[0041] Next, the derivation unit 12 derives feature data including one or more input features based on the detection results (step S2).
[0042] Next, the estimation unit 13 inputs the combination of feature data and condition data into the AI model 100, thereby obtaining the estimation result (step S3).
[0043] The output unit 14 outputs the estimation result to the external device 5 (step S4), which enables control of the traffic signal 60 or the on-board equipment of a vehicle C traveling in the vicinity.
[0044] Next, the effects and benefits of the entry estimation system 1 according to this embodiment will be described.
[0045] The entry estimation system 1 includes a camera 20 that detects a user Y on the sidewalk E2 and outputs a detection result, and a control unit 10 that estimates whether the user Y enters the crosswalk E1 from the sidewalk E2. The control unit 10 includes an acquisition unit 11 that acquires condition data including the detection result and one or more conditions related to the detection of the user Y by the camera 20, a derivation unit 12 that derives feature amount data including one or more input feature amounts based on the detection result, an estimation unit 13, and an output unit 14 that outputs an estimation result by the estimation unit 13. The estimation unit 13 obtains an estimation result by inputting the feature amount data derived by the derivation unit 12 and the condition data acquired by the acquisition unit 11 to an AI model 100 that outputs an estimation result as to whether the user Y enters the crosswalk E1 from the sidewalk E2 according to a combination of the feature amount data and the condition data.
[0046] According to such an entry estimation system 1, the feature amount data including one or more input feature amounts derived according to the detection result and the condition data acquired by the acquisition unit 11 are input to the AI model 100, and an estimation result as to whether the user Y enters the crosswalk E1 from the sidewalk E2 is output according to a combination of the feature amount data and the condition data. Thus, in the entry estimation system 1, an estimation result corresponding to a combination of the feature amount data and the condition data according to the detection result is output. Thereby, the entry estimation system 1 can output an estimation result with higher accuracy as compared with a case where an estimation result is output only from the feature amount data (for example, the trajectory of the user Y) according to the detection result, in that conditions (for example, time zone, weather, etc.) related to the detection of the user Y are taken into account. For example, when the weather is rainy, the entry estimation system 1 can estimate that the user Y enters the crosswalk E1 from a combination of the feature amount data (the trajectory of the user Y passes through the route) and the condition data (the weather is rainy) when there is a route through which the user Y entering the crosswalk E1 is likely to pass, and can perform entry estimation with high accuracy in consideration of the condition data. Therefore, according to the entry estimation system 1 according to the present embodiment, the entry estimation accuracy for a specific area can be improved.
[0047] The superiority of the entry estimation system 1 according to the present embodiment as compared with the comparative example will be described. FIGS. 8 and 9 are diagrams for explaining the entry estimation according to the comparative example. In the example shown in FIG. 8, the future position of the user is predicted and the entry estimation is performed by estimating the future trajectory of the user by learning the past trajectory of the user. In such a method, problems arise such as the inability to perform estimation in cases where the user does not move, such as a user waiting for a taxi, the difficulty of estimating from a distance, and the dependence on the surrounding environment. In this regard, in the entry estimation system 1 according to the present embodiment, the problems in the comparative example shown in FIG. 8 can be solved by considering a plurality of input feature amounts and using conditional data. In the example shown in FIG. 9, the crossing intention of the user is predicted by learning a lot of data such as images. However, a problem arises in that the processing becomes highly loaded by using image data in a time series. In this regard, in the entry estimation system 1 according to the present embodiment, instead of using the image data as it is, since the input feature amount is input to the AI model 100, the processing load can be reduced.
[0048] In the entry estimation system 1, the control unit 10 may further include a learning unit 15 that learns the AI model 100 by associating the feature amount data, the conditional data, and the estimation result. In this way, in the entry estimation system 1, by learning the feature amount data, the conditional data, and the estimation result in association with each other, the estimation accuracy of the AI model 100 can be appropriately improved.
[0049] In the entry estimation system 1, the conditional data may be data including the conditions of the time zone at the time of detection by the camera 20. Thereby, the entry estimation system 1 can perform the entry estimation with high accuracy in consideration of the characteristics for each time zone (for example, a certain route is likely to be adopted for the route of the user Y who enters the second area after evening, etc.).
[0050] In the entry estimation system 1, the condition data may include data on the month or season at the time of detection by the camera 20. This allows the entry estimation system 1 to perform entry estimation with high accuracy by taking into account seasonal characteristics (for example, a certain area may be shaded by trees and therefore more likely to be used as a waiting position for user Y entering the second area during the hot summer months).
[0051] In the entry estimation system 1, the condition data may include data on the temperature at the time of detection by the camera 20 or within a predetermined time period from the time of detection. This allows the entry estimation system 1 to perform entry estimation with high accuracy by taking into account temperature characteristics (for example, a certain area may be shaded by trees and therefore more likely to be adopted as a waiting position for user Y entering the pedestrian crossing E1 when the temperature is high).
[0052] In the entry estimation system 1, the condition data may include weather conditions at the time of detection by the camera 20 or within a predetermined time period from the time of detection. This allows the entry estimation system 1 to perform entry estimation with high accuracy by taking into account weather characteristics (for example, a certain area does not get wet in the rain, so it is likely to be adopted as a waiting position for user Y entering the pedestrian crossing E1 when it is raining).
[0053] In the entry estimation system 1, the derivation unit 12 may derive at least one of the following as input features based on the detection results: the trajectory of user Y, the type of user Y, the orientation of user Y's face, environmental data of the area where user Y is located, and the distance from user Y to the pedestrian crossing E1. This allows the entry estimation system 1 to appropriately estimate whether or not a user will enter the pedestrian crossing E1.
[0054] The various embodiments and modifications described above may be combined as appropriate without departing from the spirit of this disclosure.
[0055] 1...Entry estimation system 5...External device 10...Control unit 11...Acquisition unit 12...Derivation unit 13...Estimation unit 14...Output unit 15...Learning unit 16...Memory unit 20...Camera (detection unit) 20A...Camera 20B...Camera 50...Pole 60...Traffic light 80...Millimeter wave radar 100...AI model 120...Circuit 121...Processor 122...Memory 123...Storage 124...Input / output port C...Vehicle D1...Detection range D2...Detection range D3...Detection range E1...Crosswalk E2...Sidewalk Y...User
Claims
1. An entry estimation system comprising: a detection unit that detects a user in a first area and outputs a detection result; a control unit that estimates whether or not the user enters a second area from the first area, wherein the control unit includes: an acquisition unit that acquires condition data including the detection result and one or more conditions related to the detection of the user by the detection unit; a derivation unit that derives feature data including one or more input features based on the detection result; an estimation unit that obtains an estimation result by inputting the feature data derived by the derivation unit and the condition data acquired by the acquisition unit to an AI model that outputs an estimation result of whether or not the user enters a second area from the first area according to a combination of the feature data and the condition data; and an output unit that outputs the estimation result by the estimation unit.
2. The entry estimation system according to claim 1, wherein the control unit further comprises a learning unit that learns the AI model by associating the feature data, the condition data, and the estimation results.
3. The entry estimation system according to claim 1 or claim 2, wherein the condition data includes data on the time period at the time of detection by the detection unit.
4. The entry estimation system according to any one of claims 1 to 3, wherein the condition data includes data on the month or season at the time of detection by the detection unit.
5. The entry estimation system according to any one of claims 1 to 4, wherein the condition data includes data on the temperature at the time of detection by the detection unit or within a predetermined time period from the time of detection.
6. The approach estimation system according to any one of claims 1 to 5, wherein the condition data includes weather conditions at the time of detection by the detection unit or within a predetermined time period from the time of detection.
7. The entry estimation system according to any one of claims 1 to 6, wherein the derivation unit derives at least one of the following as the input feature quantities based on the detection result: the user's trajectory, the user's type, the orientation of the user's face, environmental data of the area where the user is located, and the distance from the user to the second area.