In-vehicle device and object detection method

Abstract

To further improve the accuracy of object detection. [Solution] The in-vehicle device according to the embodiment includes a controller that detects objects from images captured by an in-vehicle camera. The controller also detects the objects as detection frames, including the object's class and confidence level, through image recognition processing of the images, and changes the confidence level of the detection frames according to the arrangement relationship of the detection frames with respect to pre-set areas for each object class.

Description

Technical Field

[0001] The disclosed embodiments relate to in-vehicle devices and object detection methods.

Background Art

[0002] Conventionally, a technique for detecting an object around a vehicle from an image captured by an in-vehicle camera by image recognition processing using an AI (Artificial Intelligence) model has been known. The detection result of an object using such a technique is used, for example, in an advanced driver assistance system (ADAS: Advanced Driver-Assistance Systems) or the like.

[0003] By the way, when using such an object detection technique, misdetection may occur, such as an object that should originally be detected as a vehicle being detected as another object such as a traffic signal or a sign. For example, when a truck is detected as a surrounding vehicle during night driving, the taillight provided on the upper rear surface of the truck's trailer may be detected as a red signal together with the truck.

[0004] In order to prevent such misdetection, for example, a technique has been proposed in which the reliability indicating the accuracy associated with the detection frame of each object is compared, and the position, vertical width, horizontal width, etc. of the detection frame with low reliability are corrected based on the detection frame with high reliability (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] However, if the conventional technology described above is used, for example, if both the detection frame for the taillight detected as a red light and the detection frame for the truck have high reliability, it will not be possible to prevent false detections.

[0007] One embodiment has been made in view of the above, and aims to provide an in-vehicle device and an object detection method that can further improve the accuracy of object detection. [Means for solving the problem]

[0008] One embodiment includes a controller that detects objects from images captured by an in-vehicle camera. The controller detects the objects as detection frames, including the object's class and confidence level, through image recognition processing of the images, and changes the confidence level of the detection frames according to the arrangement of the detection frames within pre-set areas for each object class. [Effects of the Invention]

[0009] According to one embodiment, the controller can change the reliability of a detection frame depending on its positional relationship with a pre-defined area for each class of object. For example, if the positional relationship of the detection frame with respect to an area indicates a high probability of false detection, the reliability of the detection frame can be reduced. In other words, according to one embodiment, the accuracy of object detection can be further improved. [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 is a schematic diagram illustrating the object detection method according to an embodiment. [Figure 2] Figure 2 shows an example of the configuration of an image processing system according to an embodiment. [Figure 3] Figure 3 shows an example of the configuration of an image processing apparatus according to the embodiment. [Figure 4] Figure 4 shows an example of the configuration of an image processing apparatus according to a modified example. [Figure 5]Figure 5 is a flowchart showing the processing procedure performed by the image processing device according to the embodiment. [Figure 6] Figure 6 shows a specific example of confidence map information. [Figure 7] Figure 7 shows an example of changing the confidence level in Case 2. [Figure 8] Figure 8 shows examples of changes in confidence levels in Case 3 and Case 4. [Figure 9] Figure 9 shows examples of changes in confidence levels in Case 5 and Case 6. [Modes for carrying out the invention]

[0011] The embodiments of the in-vehicle device and object detection method disclosed herein will be described in detail below with reference to the attached drawings. However, the present invention is not limited to the embodiments described below.

[0012] Furthermore, the following examples illustrate how the in-vehicle device according to the embodiment may be implemented as a drive recorder 11 (see Figure 3) or an image processing ECU (Electronic Control Unit) 31 (see Figure 4). The object detection method according to the embodiment is an object detection method executed by an image processing device 50 of the drive recorder 11 or the image processing ECU 31.

[0013] Furthermore, the terms "designated," "specific," and "certain" in the following explanation may be interpreted as "predetermined." Also, "reliability" may be interpreted as "accuracy," which indicates the degree of precision.

[0014] First, an overview of the object detection method according to the embodiment will be explained using Figure 1. Figure 1 is an overview diagram of the object detection method according to the embodiment.

[0015] The image processing apparatus 50 according to the embodiment detects an object from an image captured by an in-vehicle camera and changes the reliability of a detection frame according to the arrangement relationship of the detection frames with respect to an area preset for each class (type) of the object. Such information processing is executed by a controller 52 (see FIGS. 3 or 4) included in the image processing apparatus 50.

[0016] Specifically, as shown in FIG. 1, the controller 52 detects each object from the image captured by the in-vehicle camera (step S1). At this time, the controller 52 detects each object using an image recognition model 51b (see FIGS. 3 or 4).

[0017] The image recognition model 51b is an AI model for image recognition. The AI model is, for example, a DNN (Deep Neural Network) model learned using a machine learning algorithm.

[0018] The image recognition model 51b operates as an image recognition AI by being loaded into the controller 52 as an AI model. When the image recognition model 51b operates as an image recognition AI, when each frame of the video captured by the in-vehicle camera is input, it is pre-learned to output the class, position, and size of various objects appearing in each frame as detection frames. Also, the image recognition model 51b is pre-learned to output the color etc. of the detected object when operating as an image recognition AI. In addition, the image recognition model 51b is pre-learned to output the reliability of each detection frame when operating as an image recognition AI.

[0019] In the present embodiment, the image recognition model 51b is learned to output at least other vehicles around the host vehicle, license plates of vehicles, signs, lanes, traffic lights, the lighting color of traffic lights, the reliability of each detection frame, etc.

[0020] However, with this image recognition model 51b, false detections can occur due to various conditions such as the image capture conditions, the position of each object relative to the vehicle, and relative speed. For example, Figure 1 shows an example where, at night, a truck traveling in an adjacent lane was detected as a "vehicle" in detection frame BB1, but the taillights on the upper rear of the truck's trailer were also falsely detected as "traffic lights" (red lights) in detection frames BB2 and BB3. Hereafter, this example shown in Figure 1 will be referred to as "Case 1".

[0021] To prevent such false detections, in the object detection method according to the embodiment, the controller 52 changes the confidence level of the detection frame according to the arrangement relationship of the detection frame with respect to a pre-set area for each class of object (step S2). At this time, the controller 52 changes the confidence level of the detection frame based on confidence level map information 51c (see Figure 3 or Figure 4), which sets the content of the confidence level change according to the arrangement relationship of the detection frame with respect to the area.

[0022] Specifically, as shown in step S2, in case 1 where a traffic light is detected within the vehicle's detection frame BB1, the controller 52 reduces the reliability of the corresponding traffic light's detection frames BB2 and BB3. This reduces the reliability of the detection result, which is unnatural and highly likely to be a false detection, where the traffic light is detected within the vehicle's detection frame, thereby improving the accuracy of object detection.

[0023] The reliability map information 51c includes, for example, a vehicle detection frame set as the corresponding area when a traffic light is detected, and the placement relationship of the traffic light detection frame within this corresponding area, such as whether it is inside or outside the vehicle detection frame, is linked to it. Furthermore, each placement relationship is linked to the details of how the reliability of the traffic light detection frame is changed.

[0024] For example, in cases where there is a high probability of false detection, such as in Case 1 where the detection frames BB2 and BB3 of the traffic signals are within the detection frame BB1 ​​of the vehicle, the reliability map information 51c is pre-set to reduce the reliability of the detection frames BB2 and BB3 of the traffic signals. A specific example of the reliability map information 51c will be explained later using Figure 6.

[0025] The controller 52 modifies the confidence level of each detection frame based on this confidence map information 51c. This allows for, for example, reducing the confidence level of detection patterns for unnatural objects that are anticipated in advance, thereby preventing false detections. The controller 52 then outputs the detection results, including the modified confidence levels, to various devices that utilize the detection results from the image processing device 50 (for example, devices that perform various ADAS functions) and the center device 100 (see Figure 2).

[0026] In the object detection method according to this embodiment, the controller 52 detects objects from images captured by the in-vehicle camera. The controller 52 also detects the objects as detection frames, including the object's class and confidence level, through image recognition processing of the images. Furthermore, the controller 52 changes the confidence level of the detection frames according to the arrangement relationship of the detection frames within pre-set areas for each object class.

[0027] Therefore, according to the object detection method according to the embodiment, the reliability of the detection frame can be changed depending on the arrangement relationship of the detected detection frame with respect to a pre-set area for each class of object. For example, if the arrangement relationship of the detection frame with respect to the area indicates a high possibility of false detection, the reliability of the detection frame can be reduced. In other words, according to the object detection method according to the embodiment, the accuracy of object detection can be further improved.

[0028] In Case 1 described above, we provided an example where the reliability of the detection frame was reduced, but in other cases, the reliability of the detection frame may be increased. Specific examples of other cases besides Case 1, including such cases, will be explained later using Figures 6 to 9.

[0029] The following describes in more detail an example of the configuration of an image processing system 1, which includes an in-vehicle device 10 having an image processing device 50 to which the object detection method according to the above embodiment is applied.

[0030] Figure 2 shows an example of the configuration of the image processing system 1 according to the embodiment. As shown in Figure 2, the image processing system 1 includes in-vehicle devices 10-1, 10-2, ..., 10-m (where m is a natural number of 3 or greater) and a center device 100.

[0031] Each in-vehicle device 10 and the central device 100 are connected to each other via a network N1, such as the Internet, a mobile phone network, or a C-V2X (Cellular Vehicle to Everything) communication network, enabling them to communicate with one another.

[0032] The central device 100 is implemented, for example, as a private cloud. The central device 100 is managed, for example, by a business operator that operates a data center that integrates various data transmitted from the in-vehicle devices 10. The central device 100 collects status data indicating the status of each vehicle transmitted from each in-vehicle device 10. The status data includes the detection results of objects detected by the image processing device 50.

[0033] Furthermore, the central device 100 analyzes the collected situation data and performs various information processing based on the analysis results. For example, the central device 100 generates driver assistance information for each vehicle based on the analysis results and transmits it to each in-vehicle device 10.

[0034] Next, an example of the configuration of the image processing device 50 in each in-vehicle device 10 will be described. Figure 3 is a diagram showing an example of the configuration of the image processing device 50 according to the embodiment. As shown in Figure 3, the image processing device 50 includes a storage unit 51 and a controller 52. The image processing device 50 is also connected to a communication unit 12, a camera 13, an output unit 14, and an external device 70.

[0035] The communication unit 12 is implemented by a network adapter or the like. The communication unit 12 is wirelessly connected to the network N1 and transmits and receives information to and from the center device 100 via the network N1.

[0036] Camera 13 is one or more on-board cameras mounted on various parts of the vehicle. Camera 13 is positioned to capture at least the area in front of the vehicle. Camera 13 may be positioned to capture not only the area in front of the vehicle, but also the area behind or to the sides of the vehicle, and even the interior of the vehicle. Camera 13 may also be a 360-degree camera capable of capturing the entire area around the vehicle.

[0037] The output unit 14 is an output device that displays output information from the image processing device 50. The output unit 14 is implemented by a display, speaker, etc. The external device 70 is various devices that utilize the detection results from the image processing device 50 (as described above, for example, devices that perform various ADAS functions).

[0038] The image processing device 50 is, for example, a computer mounted in a vehicle. The image processing device 50 performs at least the information processing steps S1 and S2 described with reference to Figure 1.

[0039] The memory unit 51 is implemented by a memory device such as RAM (Random Access Memory) or flash memory. The memory unit 51 stores a program (not shown) according to an embodiment executed by the controller 52. The memory unit 51 also stores various types of information used in the information processing executed by the controller 52.

[0040] In the example shown in Figure 3, the memory unit 51 stores various types of information, including camera setting information 51a, an image recognition model 51b, and confidence map information 51c. The camera setting information 51a is information related to the settings of the camera 13. The camera setting information 51a is set when the camera 13 is installed, etc. The camera setting information 51a includes the position of the horizontal line in the image, the set position of the vehicle body indicating the position of the hood, etc.

[0041] The image recognition model 51b has already been explained, so its explanation will be omitted here. The confidence map information 51c is map information that sets how the confidence level of the detection frame is changed for each class of object indicated by the detected detection frame, depending on the arrangement of the detected detection frame. A specific example of the confidence map information 51c will be explained later using Figure 6. Note that the latest versions of the image recognition model 51b and the confidence map information 51c may be distributed from the center device 100 as needed.

[0042] The controller 52 corresponds to a so-called processor. The controller 52 can be implemented by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), etc. The controller 52 reads the program according to the embodiment stored in the memory unit 51 and executes it using RAM as the working area. The controller 52 can also be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0043] The controller 52 performs information processing according to the processing procedure shown in Figure 5. An explanation using Figure 5 will be given later.

[0044] As shown in Figure 3, the in-vehicle device 10 can be implemented as a drive recorder 11 having a communication unit 12, a camera 13, an output unit 14, and an image processing device 50. In this case, the camera 13 is implemented by a camera unit mounted on the drive recorder 11. The output unit 14 is implemented by a display and speaker mounted on the drive recorder 11. The image processing device 50 is implemented by a microcontroller mounted on the drive recorder 11.

[0045] In the example shown in Figure 3, the external device 70 is connected to the image processing device 50 via an in-vehicle network such as CAN (Controller Area Network). The external device 70 may also be connected to the image processing device 50 wirelessly via Bluetooth®, Wi-Fi®, UWB (Ultra Wide Band), or the like.

[0046] Furthermore, the configuration example shown in Figure 3 is just one example, and other modifications can be given. Figure 4 is a diagram showing a configuration example of the image processing device 50 according to a modification. As shown in Figure 4, the in-vehicle device 10 can implement the image processing device 50 as an image processing ECU 31.

[0047] In this case, the communication unit 12 is implemented by, for example, a network adapter mounted on the drive recorder 11. The camera 13 is implemented by, for example, a camera unit mounted on the drive recorder 11. The camera 13 may also be implemented by an in-vehicle camera unit other than the drive recorder 11. The output unit 14 is implemented by an in-vehicle output device 21 mounted on the vehicle, such as an in-vehicle display or in-vehicle speaker.

[0048] In the example shown in Figure 4, the communication unit 12, camera 13, output unit 14, and external device 70 are connected to the image processing device 50 via an in-vehicle network such as CAN. The communication unit 12, camera 13, output unit 14, and external device 70 may also be connected to the image processing device 50 wirelessly via Bluetooth®, Wi-Fi®, UWB (Ultra Wide Band), etc.

[0049] Next, the processing procedure for information processing performed by the controller 52 of the image processing device 50 will be explained using Figure 5. Figure 5 is a flowchart showing the processing procedure performed by the image processing device 50 according to this embodiment.

[0050] First, the controller 52 acquires an image from the camera 13 (step S101). Then, the controller 52 performs image recognition processing on the acquired image using the image recognition model 51b to detect an object (step S102).

[0051] Next, the controller 52 obtains one detection frame from the detection results in step S102 (step S103). Then, the controller 52 compares the obtained detection frame with the confidence map information 51c (step S104) and determines whether or not to change the confidence level based on the comparison result (step S105).

[0052] If a change in confidence level is required (step S105, Yes), the controller 52 changes the confidence level of the detection frame based on the confidence level map information 51c (step S106) and proceeds to step S107. If a change in confidence level is not required (step S105, No), the controller 52 proceeds to step S107 without changing the confidence level.

[0053] Here, we will explain the process of changing the confidence level in cases other than Case 1 described above, referring to a specific example of the confidence map information 51c. Figure 6 is a diagram showing a specific example of the confidence map information 51c.

[0054] Figure 7 shows an example of changing the confidence level in Case 2. Figure 8 shows an example of changing the confidence level in Cases 3 and 4. Figure 9 shows an example of changing the confidence level in Cases 5 and 6. Of the Cases 1 to 7 shown in Figures 6 to 9, Cases 1 to 6 correspond to cases with a high probability of false positives.

[0055] As shown in Figure 6, the confidence map information 51c is map information that sets how the confidence level of the detection frame is changed depending on the arrangement of the detected detection frames in the area corresponding to each class of object indicated by the detected detection frame.

[0056] As shown in Figure 6, for example, Case 1, which has already been explained, is the case where the class of the detection frame to be judged is "traffic light" (ID: 01), and that detection frame is "inside" the corresponding area, the "vehicle detection frame" (see "*Case 1" in the figure). In this case, because the reliability of the reliability map information 51c is set to "decreased", the controller 52 decreases the reliability of the corresponding "traffic light" detection frame.

[0057] For example, in Case 2, the class of the detection frame to be judged is also "traffic light," and that detection frame "overlaps" with the corresponding area, which is "detection frame of another traffic light" (see "*Case 2" in the figure). In other words, this is the case when one traffic light is detected multiple times. In this case, because the reliability of the reliability map information 51c is set to "decreased," the controller 52 decreases the reliability of the detection frame of the corresponding "traffic light."

[0058] Specifically, as shown in Figure 7, if the controller 52 detects a single traffic light multiple times in Case 2, it reduces the reliability of the detection frames BB11 and BB12 for the corresponding "traffic light". At this time, the controller 52 may reduce the reliability of both detection frames BB11 and BB12, or it may reduce the reliability of only one of them.

[0059] If the reliability of only one of the detection frames is to be reduced, the controller 52 will, for example, prioritize the detection frame with higher reliability and further reduce the reliability of the detection frame with lower reliability. In this case, the controller 52 may increase the reliability of the detection frame with higher reliability instead of reducing the reliability of the detection frame with lower reliability. This reduces the reliability of detection results that are unnatural and have a high probability of false detection, such as when a single traffic light is detected multiple times, and further improves the accuracy of object detection.

[0060] Let's return to the explanation of Figure 6. For example, Case 3 is the case where the class of the detection frame to be judged is also "traffic light", and that detection frame is "inside" the corresponding area, "area below the horizontal line" (see "*Case 3" in the figure). In other words, this is the case when a traffic light is detected below the horizontal line. In this case, because the reliability of the reliability map information 51c is set to "decreased", the controller 52 decreases the reliability of the detection frame for the corresponding "traffic light".

[0061] For example, in Case 4, the class of the detection frame to be judged is "vehicle," and that detection frame is "inside" the corresponding area, which is "the area above the horizontal line" (see "*Case 4" in the figure). In other words, this is the case when a vehicle is detected above the horizontal line. In this case, because the confidence level of the confidence level map information 51c is set to "decreased," the controller 52 decreases the confidence level of the corresponding "vehicle" detection frame.

[0062] Specifically, as shown in Figure 8, if the controller 52 detects a traffic light below the horizontal line position HL1 in case 3, it reduces the reliability of the detection frame BB21 for the corresponding "traffic light". This reduces the reliability of the detection result, which is unnatural and likely to be a false detection, where the traffic light is detected below the horizontal line position HL1, thereby improving the accuracy of object detection. Also, if the controller 52 detects a vehicle above the horizontal line position HL1 in case 4, it reduces the reliability of the detection frame BB22 for the corresponding "vehicle". This reduces the reliability of the detection result, which is unnatural and likely to be a false detection, where the vehicle is detected above the horizontal line position HL1, thereby improving the accuracy of object detection. Note that the controller 52 refers to the camera setting information 51a for the horizontal line position HL1.

[0063] Let's return to the explanation of Figure 6. For example, Case 5 is when the class of the detection frame to be judged is "traffic light" or "vehicle," and that detection frame is "inside" the corresponding area, which is "the area below the set position of the vehicle body" (see "*Case 5" in the figure). In other words, this is when a traffic light or vehicle is detected below the set position of the vehicle body. In this case, because the reliability of the reliability map information 51c is set to "decreased," the controller 52 decreases the reliability of the detection frame for the corresponding "traffic light" or "vehicle."

[0064] For example, in Case 6, the class of the detection frame to be judged is "sign," and that detection frame is "inside" the corresponding area, which is the "vehicle lane" (see "*Case 6" in the diagram). In other words, this is the case when a sign is detected inside the vehicle lane. In this case, because the reliability of the reliability map information 51c is set to "decreased," the controller 52 decreases the reliability of the detection frame for the corresponding "sign."

[0065] Specifically, as shown in Figure 9, if the controller 52 detects a traffic light or vehicle below the set position SP1 on the vehicle body (Case 5), it reduces the reliability of the detection frame BB31 for the corresponding "traffic light" or "vehicle." This reduces the reliability of the detection result, which is unnatural and highly likely to be a false detection, such as when a traffic light or vehicle is detected below the set position SP1 on the vehicle body, thereby improving the accuracy of object detection. The controller 52 refers to the camera setting information 51a for the set position SP1 on the vehicle body.

[0066] Furthermore, if the controller 52 detects a sign within its own lane (Case 6), it reduces the reliability of the detection frame BB32 for the corresponding "sign." On the other hand, the controller 52 does not change the reliability of the detection frame BB33 for "signs" outside its own lane. This reduces the reliability of detection results that are unnatural and highly likely to be false detections, such as when a sign is detected within the vehicle's lane, thereby improving the accuracy of object detection. The controller 52 estimates the vehicle's lane from the image-recognized lanes L1 and L2.

[0067] Let's return to the explanation of Figure 6. For example, Case 7 is the case where the class of the detection frame to be judged is "vehicle," and that detection frame "encompasses" the corresponding area, the "license plate detection frame" (see "*Case 7" in the figure). In other words, a vehicle is detected, and the license plate is detected within the vehicle's detection frame. In this case, the vehicle's detection frame is highly likely to actually be a vehicle (a correct detection). In this case, because the confidence level of the confidence level map information 51c is set to "increase," the controller 52 increases the confidence level of the corresponding "vehicle" detection frame. This increases the confidence level of the detection result, which is highly likely to be a correct detection because the vehicle's detection frame encompasses the license plate, and further improves the accuracy of object detection.

[0068] In Figure 6, the confidence map information 51c shows an example where "decrease" or "increase" is set as the change in confidence level, but this does not limit the values ​​that can be set for such changes. Therefore, for example, specific increase or decrease values ​​for confidence level, or updated confidence levels (the changed values) may be set as the change.

[0069] Returning to the explanation of Figure 5, the controller 52 then determines whether there are any undetermined detection frames among the detection frames detected in step S102 (step S107). If there are undetermined detection frames (step S107, Yes), the controller 52 repeats the process from step S103.

[0070] If there are no undetermined detection frames (step S107, No), the controller 52 outputs the object detection results, including the revised reliability, to the external device 70 and the center device 100 (step S108). When the controller 52 uses the object detection results itself, it appropriately performs the necessary information processing based on the object detection results and outputs the processing results to the output unit 14.

[0071] The controller 52 then determines whether or not the system has terminated (step S109). The controller 52 determines that the system has terminated if, for example, the ignition switch is turned off.

[0072] If the system is not shutting down (step S109, No), the controller 52 repeats the process from step S101. If the system is shutting down (step S109, Yes), the controller 52 terminates the process.

[0073] As described above, the in-vehicle device 10 according to the embodiment includes a controller 52 that detects objects from images captured by a camera 13 (corresponding to an example of an "in-vehicle camera"). The controller 52 also detects the objects as detection frames that include the class and confidence level of the objects through image recognition processing of the images, and changes the confidence level of the detection frames according to the arrangement relationship of the detection frames with respect to pre-set areas for each class of object.

[0074] Therefore, according to the in-vehicle device 10 of the embodiment, the reliability of the detection frame can be changed depending on the arrangement relationship of the detected detection frame with respect to a pre-set area for each class of object. For example, if the arrangement relationship of the detection frame with respect to the area indicates a high possibility of false detection, the reliability of the detection frame can be reduced. In other words, according to the in-vehicle device 10 of the embodiment, the accuracy of object detection can be further improved.

[0075] In the embodiments described above, cases 1 to 7 were given as examples in the explanation of how to change the reliability level, but this does not limit the cases in which a change in reliability level is necessary. Therefore, taking Figure 6 as an example, the reliability map information 51c may further include the arrangement relationships and changes in reliability level for areas corresponding to each class of object other than "traffic lights," "vehicles," and "signs."

[0076] Furthermore, in the above-described embodiment, the controller 52 of the image processing device 50 in the in-vehicle device 10 is responsible for executing the object detection method according to the embodiment, but the controller in the center device 100 may also perform the execution. In this case, the center device 100 has an AI model corresponding to the image recognition model 51b and map information corresponding to the confidence map information 51c, and uses these to detect objects from images collected from each in-vehicle device 10.

[0077] Further effects and modifications can be readily derived by those skilled in the art. Therefore, broader aspects of the present invention are not limited to the specific details and representative embodiments expressed and described above. Accordingly, various modifications are possible without departing from the spirit or scope of the overall concept of the invention as defined by the appended claims and their equivalents. [Explanation of Symbols]

[0078] 1. Image Processing System 10 Onboard equipment 11. Dashcam 12 Communications Department 13 Cameras 14 Output section 21. On-board output device 31 Image Processing ECU 50 Image Processing Devices 51 Storage section 51a Camera settings information 51b Image Recognition Model 51c Confidence Map Information 52 Controllers 70 External device 100 Center device

Claims

1. It is equipped with a controller that detects objects from images captured by the in-vehicle camera, The aforementioned controller, Image recognition processing on the aforementioned image detects the object as a detection frame including the object's class and confidence level. The reliability of the detection frame is changed according to the arrangement of the detection frame in a pre-defined area for each class of object. In-vehicle device.

2. The aforementioned controller, The reliability of the detection frame is changed based on reliability map information, which sets the reliability changes according to the arrangement relationship of the detection frame in the area. The in-vehicle device according to claim 1.

3. The aforementioned controller, If the detection frame is positioned in a way that makes false detections highly likely in relation to the area, the reliability of the detection frame is reduced. When the detection frame is positioned in a way that makes it highly likely to detect a positive result in the area, the reliability of the detection frame is increased. The in-vehicle device according to claim 1 or 2.

4. The aforementioned controller, If the class of the detection frame is a traffic light, and the detection frame of the traffic light is within the detection frame of a vehicle, the reliability of the detection frame of the traffic light is reduced. The in-vehicle device according to claim 3.

5. The aforementioned controller, If the class of the detection frame is a traffic light, and the detection frame of the traffic light overlaps with the detection frame of another traffic light, the reliability of at least one of the traffic light and the other traffic light is reduced. The in-vehicle device according to claim 3.

6. The aforementioned controller, If the class of the detection frame is a traffic light, and the detection frame of the traffic light is below the horizontal line, the reliability of the detection frame of the traffic light is reduced. The in-vehicle device according to claim 3.

7. The aforementioned controller, If the class of the detection frame is a vehicle, and the detection frame of the vehicle is above the horizontal line, the reliability of the detection frame of the vehicle is reduced. The in-vehicle device according to claim 3.

8. The aforementioned controller, If the class of the detection frame is a traffic light or a vehicle, and the detection frame for the traffic light or vehicle is below a predetermined position on the vehicle body, the reliability of the detection frame for the traffic light or vehicle is reduced. The in-vehicle device according to claim 3.

9. The aforementioned controller, If the class of the detection frame is a sign, and the detection frame of the sign is within the vehicle's lane, the reliability of the detection frame of the sign is reduced. The in-vehicle device according to claim 3.

10. The aforementioned controller, When the class of the detection frame is a vehicle, and the detection frame of the vehicle includes a license plate detection frame, the reliability of the detection frame of the vehicle is increased. The in-vehicle device according to claim 3.

11. A method for detecting an object that is performed by a controller, Detecting objects from images captured by an in-car camera, By performing image recognition processing on the aforementioned image, the object is detected as a detection frame including the object's class and confidence level, The reliability of the detection frame is changed according to the arrangement relationship of the detection frame with respect to a pre-set area for each class of the object, An object detection method that includes [a specific method].

Images