Recognition system and object recognition method

Abstract

This recognition system for recognizing an object is characterized by comprising a calculation device which executes a program and a storage device which the calculation device can access, and is characterized in that: the calculation device has an object detection processing unit for outputting, using a detection model, a detection result obtained by extracting a region where an object to be extracted may be present from inputted image data or point group data, together with a score indicating the degree of certainty of the detection result; the calculation device has a recognition determination processing unit for generating a recognition result on the basis of a result of evaluating the score of the detection result by a predetermined threshold value; the calculation device has an output processing unit for outputting the recognition result; and the recognition determination processing unit changes the predetermined threshold value in accordance with a parameter concerning the detection result.

Description

[Technical Field]

[0001] The present invention relates to a recognition system for recognizing an object from observation data, and more particularly to a method for recognizing an object. [Background technology]

[0002] To realize self-driving cars, surrounding objects such as vehicles and pedestrians are detected and recognized from image data acquired by cameras and point cloud data acquired by LiDAR, and their positions and movements are ascertained. Since the likelihood of an object varies depending on its position and appearance, the recognition process recognizes object candidates whose likelihood is above a certain threshold. However, if the same threshold is used for all objects, there is a risk that objects with low likelihood, such as those located far away or partially hidden, may not be recognized.

[0003] The following prior art exists as background art in this technical field. Patent Document 1 (JP 2021-18465 A) describes an object recognition device that includes a storage device and a processor, and stores peripheral information and tolerance information in the storage device. The tolerance information is information that indicates the tolerance for non-detection of an object for each class of the object. The peripheral information is acquired by a sensor device provided in the vehicle. The processor performs object recognition processing to recognize objects around the vehicle. In the object recognition processing, the processor identifies the object to be detected and its class based on the peripheral information, and calculates likelihood, which is a parameter that indicates the probability of object detection. The object recognition device also describes an object recognition device in which the processor calculates a threshold value for the likelihood corresponding to the object based on the tolerance information, and determines whether to output an object identification result based on a comparison between the likelihood and the threshold value. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-18465 Summary of the Invention [Problem to be solved by the invention]

[0005] In the above-mentioned Patent Document 1, in the object recognition process that identifies the type of detected object based on peripheral information and calculates a likelihood that indicates the accuracy of object detection, a likelihood threshold corresponding to the object is calculated based on tolerance information, and the same threshold is used for the same object. Generally, when an object is far away or is partially hidden, the accuracy is low, making it difficult to recognize the object.

[0006] Therefore, the present invention aims to improve the recognition rate by changing the threshold value depending on information other than the likelihood, such as the distance to the object or the size of the object, when the object to be recognized is located far away or part of the object is hidden, and by making a judgment that makes an object that is difficult to recognize easier to recognize. [Means for solving the problem]

[0007] A representative example of the invention disclosed in the present application is as follows: That is, a recognition system for recognizing an object, comprising: a computing device that executes a program; and a storage device accessible by the computing device, wherein the computing device uses a detection model to extract an area where an object to be extracted from input image data or point cloud data may exist, and outputs a detection result obtained by extracting an area together with a score representing the likelihood of the detection result; a recognition determination processing unit that generates a recognition result based on a result of evaluating the score of the detection result using a predetermined threshold; and an output processing unit that outputs the recognition result; Depending on the distance to the object, the image is displayed by a quadratic function with little change in the vicinity. The predetermined threshold value is changed. [Effects of the Invention]

[0008] According to one aspect of the present invention, it is possible to improve the recognition rate of an object. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing the configuration of a recognition device that recognizes an object using image data acquired by a camera. [Figure 2] FIG. 1 is a block diagram showing the configuration of a recognition device that recognizes an object using point cloud data acquired by a LiDAR. [Figure 3] 10 is a flowchart of an operation when image data acquired by a camera is used. [Figure 4] 10 is a flowchart of an operation when point cloud data acquired by LiDAR is used. [Figure 5] FIG. 10 is a diagram illustrating an example of an image. [Figure 6A] FIG. 10 is a diagram showing output data from an object detection processing unit. [Figure 6B] FIG. 10 is a diagram showing output data from an object detection processing unit. [Figure 7A] FIG. 10 is a diagram illustrating an example of a threshold value for recognizing image data. [Figure 7B] FIG. 10 is a diagram illustrating an example of a threshold value for recognizing image data. [Figure 8] FIG. 10 is a diagram illustrating an example of point cloud data. [Figure 9] FIG. 10 is a diagram showing output data from an object detection processing unit. [Figure 10A] FIG. 10 is a diagram illustrating an example of a threshold value for recognizing point cloud data. [Figure 10B] FIG. 10 is a diagram illustrating an example of a threshold value for recognizing point cloud data. [Figure 11] FIG. 10 is a diagram illustrating an example of thresholds defined in steps according to distance ranges. [Figure 12] FIG. 10 is a diagram illustrating an example of a threshold defined by a function of distance. [Figure 13] FIG. 10 is a diagram illustrating an example of a threshold defined by the magnitude of a detection result. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0011] Example 1 FIG. 1 is a block diagram showing the configuration of a recognition system that recognizes an object using image data acquired by a camera.

[0012] 1 includes an image data processing unit 102, an object detection processing unit 103, a recognition determination processing unit 104, a threshold table 105, and an output processing unit 106. This recognition system is mounted on a vehicle, for example, and incorporated into an electronic control unit that recognizes objects based on observation results outside the vehicle.

[0013] The electronic control device into which the recognition system is incorporated has an arithmetic unit, a storage device, and a communication interface. The arithmetic unit is a processor (e.g., a CPU) that executes programs stored in the storage device. The arithmetic unit operates as a functional unit that provides various functions by executing predetermined programs. The storage device includes a non-volatile storage area and a volatile storage area. The non-volatile storage area includes a program area that stores programs executed by the arithmetic unit, and a data area that temporarily stores data used by the arithmetic unit when executing the programs. The communication interface connects to other electronic control devices via a network such as CAN or Ethernet.

[0014] The image data processing unit 102 receives image data captured by the camera 101, adjusts the brightness and contrast of the received image data, extracts edges, and performs other processing, and then transmits the processed image data to the object detection processing unit 103. The camera 101 may be a camera that captures planar images or a depth camera that captures distance images accompanied by distance data.

[0015] The object detection processing unit 103 uses an AI model to perform recognition processing to extract areas from the image data where there is a possibility that an object to be extracted exists, calculates the degree of certainty that the extracted area is the target object, the size of the extracted area, the distance to the object, etc., and transmits the detection results to the recognition judgment processing unit 104.

[0016] The recognition determination processing unit 104 selects a threshold value for determining the certainty level received from the object detection processing unit 103 in accordance with a set value stored in a threshold value table 105, generates a recognition result when the certainty level exceeds the threshold value, and transmits the recognition result to the output processing unit 106. The recognition determination processing unit 204 also refers to the threshold value table 205 using the score output by the object detection processing unit 203 to determine a threshold value to be compared with the certainty level. As will be described later, this score can use the distance to the extracted object, the size of the object area (the number of pixels in the object area in the image data), the size of the object, etc.

[0017] The output processing unit 106 outputs the recognition result to a display device in a manner that is visible to other electronic control devices or the user.

[0018] Figure 2 is a block diagram showing the configuration of a recognition system that recognizes objects using point cloud data acquired by LiDAR. This recognition system is, for example, similar to the recognition system shown in Figure 1, mounted on a vehicle and incorporated into an electronic control unit that recognizes objects from observation results outside the vehicle. The configuration of the electronic control unit (arithmetic unit, storage device, communication interface) is also the same as described above.

[0019] The recognition system shown in FIG. 2 includes a point cloud data processing unit 202 , an object detection processing unit 203 , a recognition determination processing unit 204 , a threshold table 205 , and an output processing unit 206 .

[0020] The point cloud data processing unit 202 receives the point cloud data acquired by the LiDAR 201, performs processing such as converting the point cloud angle and distance data contained in the received point cloud data into a Cartesian coordinate system, and then transmits the processed point cloud data to the object detection processing unit 203.

[0021] The object detection processing unit 203 uses an AI model to perform detection processing to extract areas from the point cloud data where there is a possibility that an object to be extracted exists, calculates a score indicating the degree of certainty that the extracted area is the target object, the size of the extracted area, the distance to the object, etc., and transmits the detection results to the recognition judgment processing unit 204.

[0022] The recognition determination processing unit 204 selects a threshold value for determining the score received from the object detection processing unit 203 according to a set value stored in a threshold value table 205, generates a recognition result when the score exceeds the threshold value, and transmits the recognition result to the output processing unit 206. The recognition determination processing unit 204 also refers to the threshold value table 205 using the parameters output by the object detection processing unit 203 to determine a threshold value to be compared with the confidence level. As will be described later, the parameters can include the distance to the extracted object, the size of the object region (the number of points in the object region in the point cloud data), and the size of the object.

[0023] The output processing unit 206 outputs the recognition result to a display device in a manner that is visible to other electronic control devices or the user.

[0024] FIG. 3 is a flowchart of the operation when image data acquired by the camera 101 is used.

[0025] Step 301 : The image data processing unit 102 acquires one frame of image data from the camera 101 .

[0026] Step 302: The image data processing unit 102 adjusts the brightness and contrast of the acquired image data, and executes correction and processing such as edge extraction.

[0027] Step 303: The object detection processing unit 103 uses an AI model to perform recognition processing to extract areas from the image data where an object to be extracted may exist, and calculates the degree of certainty that the extracted area is the target object, the distance to the object, the size of the extracted object area (the number of pixels in the object area in the image data), the size of the object, etc. The data output from the object detection processing unit 103 in step 303 is referred to as the "detection result."

[0028] Step 304: The recognition determination processing unit 104 obtains a threshold value from the threshold value table 105 according to the attribute (for example, distance) of the detected region.

[0029] Step 305: The recognition determination processing unit 104 compares the certainty of the target object with a threshold value obtained from the threshold value table 105, and if the certainty is equal to or greater than the threshold value, it determines that the region is the target object, and proceeds to step 306. On the other hand, if the certainty is less than the threshold value, it determines that the region is not the target object, skips step 306, and ends the loop.

[0030] Step 306: The output processing unit 106 outputs information about the detected target object, such as the confidence level, the distance to the object, the size of the extracted object area, the size of the object, etc. The data output from the recognition determination processing unit 104 in step 306 is referred to as the "recognition result."

[0031] The processes from step 304 to step 306 are repeatedly executed for each candidate region of the object detected in step 303.

[0032] FIG. 4 is a flowchart of the operation when point cloud data acquired by the LiDAR 201 is used.

[0033] Step 401: The point cloud data processing unit 202 acquires one frame of point cloud data from the LiDAR 201.

[0034] Step 402: The point cloud data processing unit 202 performs processing such as converting the angle and distance data of the point cloud included in the acquired point cloud data into an orthogonal coordinate system.

[0035] Step 403: The object detection processing unit 203 uses an AI model to perform recognition processing to extract areas from the point cloud data where an object to be extracted may exist, and calculates the degree of certainty that the extracted area is the target object, the distance to the extracted object, the size of the object area (the number of points in the object area in the point cloud data), the size of the object, etc. The data output from the object detection processing unit 203 in step 303 is referred to as the "detection result."

[0036] Step 404: The recognition determination processing unit 204 obtains a threshold value from the threshold value table 205 according to the attribute (for example, distance) of the detected region.

[0037] Step 405: The recognition determination processing unit 204 compares the certainty of the target object with a threshold value obtained from the threshold value table 205, and if the certainty is equal to or greater than the threshold value, it determines that the region is the target object, and proceeds to step 406. On the other hand, if the certainty is less than the threshold value, it determines that the region is not the target object, skips step 406, and ends the loop.

[0038] Step 406: The output processing unit 206 outputs information about the detected target object, such as the confidence level, the distance to the object, the size of the extracted object area (the number of pixels in the object area in the image data), the size of the object, etc. The data output from the recognition determination processing unit 104 in step 406 is referred to as the "recognition result."

[0039] The processes from step 404 to step 406 are repeatedly executed for each candidate region of the object detected in step 403.

[0040] The distance to the object may be expressed as an X coordinate (front-rear direction) value instead of the actual distance.

[0041] Although the example of determining the distance to the target object has been described as a parameter, other parameters such as the ratio of the object size to a standard size, or the ratio of length, width, and height may also be used, or two or more parameters may be used in combination. When two or more parameters are used in combination, the parameter with the largest or smallest threshold may be used, or the recognition result may be output when all or multiple parameters are met, or the difference from the threshold (narrow loss rate) may also be taken into consideration for the determination.

[0042] Next, a specific method for comparing with the threshold value will be described.

[0043] FIG. 5 is a diagram showing an example of an image, FIG. 6A is a diagram showing output data from object detection processing unit 103, and FIG. 7A is a diagram showing an example of threshold values ​​for recognizing image data.

[0044] 5, three vehicles 501, 502, and 503 are recognized and output from the object detection processing unit 103. In this case, the data shown in FIG. 6A is output from the object detection processing unit 103. Note that the data of ID1, 2, and 3 in FIG. 6A represent the vehicles 501, 502, and 503 in FIG. 5, respectively.

[0045] In this case, if the threshold value is fixed at 0.8, vehicles 501 and 502 will be output as the recognition results, and vehicle 503 at a distance will not be recognized.

[0046] Generally, when the size of the vehicle to be detected is the same, the confidence level (score) tends to decrease as the distance increases and the number of pixels on the screen decreases, and if all detection results are judged using the same threshold value, objects at long distances will be difficult to detect. For this reason, for example, a threshold value according to distance is set, as shown in Figure 7A.

[0047] When the output data of Figure 6A is judged using the threshold value of Figure 7A, the distance to vehicle 503 with ID 3 is 110 m, and the judgment is made using a threshold value of 0.5, which is 100 m or more.Therefore, the score of vehicle 503 with ID 3, 0.55, is judged to exceed the threshold value, and the object detection processing unit 103 outputs that it has been recognized as an object.

[0048] As shown in Figures 6A and 7A, the threshold value may be determined using the distance to the object output as the detection result, or may be determined using the number of points in the object region in the image data output as the detection result, as shown in Figures 6B and 7B. Furthermore, the threshold value may be determined using the size of the object output as the detection result.

[0049] 7A and 7B, the threshold may be changed for each type of object. For example, since failing to recognize a person can cause a serious accident, the threshold may be set small for a person to increase the possibility of recognition.

[0050] Next, we will explain a specific comparison method with the threshold when using LiDAR.

[0051] FIG. 8 is a diagram showing an example of point cloud data, FIG. 9 is a diagram showing output data from object detection processing unit 203, and FIG. 10A is a diagram showing an example of thresholds for recognizing point cloud data.

[0052] In Fig. 8, three vehicles 801, 802, and 803 are recognized and output from the object detection processing unit 203. In this case, the data shown in Fig. 9 is output from the object detection processing unit 203. Note that the data of ID1, 2, and 3 in Fig. 9 represent the vehicles 801, 802, and 803 in Fig. 8, respectively.

[0053] In this case, if the threshold value is fixed at 0.8, vehicles 801 and 802 will be output as the recognition results, and vehicle 803 at a distance will not be recognized.

[0054] Generally, when the size of the vehicle to be detected is the same, the score tends to decrease as the distance increases and the number of points representing the object decreases, and if all detection results are judged using the same threshold value, objects at long distances will be difficult to detect. For this reason, for example, a threshold value according to distance is set, as shown in Figure 10A.

[0055] When the output data of Figure 9 is judged using the threshold value of Figure 10A, the distance to vehicle 803 with ID 3 is 190 m, and the judgment is made using a threshold value of 0.5, which is 150 m or more.Therefore, the score of vehicle 803 with ID 3, 0.52, is judged to exceed the threshold value, and the object detection processing unit 103 outputs that it has been recognized as an object.

[0056] As shown in Figures 9 and 10A, the threshold value may be determined using the distance to the object output as the detection result, or the X and Y coordinates of the position of the object output as the detection result may be determined as shown in Figure 10B. Furthermore, the threshold value may be determined using the number of points constituting the detected object or the size of the object region in the point cloud data output as the detection result.

[0057] Whether using image data or point cloud data, the smaller the threshold, the more recognition results are recognized as correct, but the more likely it is that a recognition result that is not actually an object will be mistakenly recognized as correct. Therefore, for each distance, a statistical value of the scores of recognition results that were correctly recognized and a statistical value of the scores of recognition results that were not correctly recognized are calculated, and a threshold is set between the two calculated statistical values. It is best to use the average value as the statistical value, but other statistical values ​​such as the median value or a combination of the minimum and maximum values ​​may also be used.

[0058] Even if the threshold is lowered and false positives increase, false positives can be eliminated by integrating the results with those of other sensors in subsequent processing, or by using the recognition results of the previous frame. However, as with conventional methods, if the recognition result is not output below the threshold, the recognition result cannot be restored in subsequent processing. Therefore, it is advisable to set the threshold low so that the number of outputs increases even if the number of false positives increases.

[0059] Furthermore, as shown in FIG. 11, the distance may be divided into multiple ranges, and the same threshold value may be used within each range. The thresholds shown in FIG. 11 have the advantage of being easy to define and easy to process. A mathematical formula for deriving a threshold value according to distance may also be used for calculation. For example, as shown in FIG. 12, if the threshold value near the vehicle position is 0.8 and the threshold value at a point 200 m away is 0.4, a linear function such as threshold line 1201 may be used, e.g., (threshold) = 0.8 - (distance) ÷ 500, or a quadratic function such as threshold line 1202 may be used, e.g., (threshold) = 0.8 - (distance) × (distance) ÷ 100,000. The slope of threshold line 1201 shown in FIG. 12 may be changed for each section (e.g., to reduce the change in the vicinity). It is also preferable that the change in the vicinity is reduced, as shown in threshold line 1202. The threshold shown in FIG. 11 may or may not be recognized near the change point, that is, the recognized object may appear and disappear, but the threshold shown in FIG. 12 changes continuously, so there is no inconvenience of the recognized object appearing and disappearing.

[0060] The same calculation method may be applied to all distances, or different calculation methods may be used for near and far distances.

[0061] Furthermore, since small objects such as fallen debris are more difficult to detect than larger objects such as vehicles at the same distance, the threshold value may be changed depending on the target object. For example, if passenger cars are used as the standard, it is advisable to set a high threshold for trucks, which are easier to detect, and a low threshold for fallen objects, which are harder to detect.

[0062] Furthermore, even if the same object is not entirely visible and part of it is hidden by another object, it may be detected as smaller than it actually is. Therefore, it is advisable to change the threshold value according to the ratio of the object's area (width x height) on the screen to the standard size when the entire object is visible. For example, as shown in FIG. 13, it is advisable to reduce the threshold value when the ratio of the visible area is small. The threshold values ​​shown in FIG. 13 may be a threshold value determined solely by the area ratio, a threshold value determined by distance (e.g., a small threshold value) used in combination, or a threshold value obtained by multiplying the threshold value determined by the area ratio by the threshold value determined by distance. In this case, it is advisable to set a lower limit for the multiplied threshold value and perform calculations to prevent the threshold value from becoming extremely small.

[0063] For point cloud data, it is advisable to set the threshold according to the ratio of the volume including the depth (the X value in Figure 9) which represents the length of the vehicle, rather than the area. Also, since it may be difficult to obtain depth information depending on the position and angle from which the object is viewed, the threshold can also be determined according to the ratio of the area (width x height) without using the depth length, even for point cloud data.

[0064] It is also possible to adjust the threshold value to be lowered by evaluating the proportion of insufficiency in part or all of the width, height, and depth, without using the area or volume of the object.

[0065] Furthermore, while vehicles are unlikely to change shape, people such as pedestrians change shape as their limbs move, making accurate recognition difficult and often resulting in low scores being output. For this reason, it is advisable to set a lower threshold for objects whose shapes change than for vehicles.

[0066] Since the recognition process is performed on all input data, it can also detect vehicles parked in roadside parking lots, for example. Vehicles that are adjacent to the road and have the potential to enter a lane should be detected, but vehicles that are located outside the guardrail and cannot enter a lane, for example, are not in the drivable area and do not need to be output as recognition results. Therefore, it is possible to use map information or the like to determine which vehicles are off the road and exclude them regardless of score, distance, or size information. Similarly, it is also advisable to exclude vehicles in the opposite lane separated by a median strip.

[0067] When calculating the score by integrating it with the detection results of other sensors in later processing, if an object at a long distance or an object that is only partially visible is detected with a low score value, the score itself may be corrected by adding a predetermined value or multiplying it by a predetermined coefficient to increase the score value.

[0068] Furthermore, under conditions that make recognition more difficult than usual due to weather factors such as rain or dirt on the sensor surface, the overall score may decrease because detection cannot be performed at the same level as usual. For this reason, the average score of all detected object candidates may be calculated for each frame, and if the calculated average is smaller than a predetermined value, the threshold may be lowered to make recognition easier.

[0069] If there are no nearby vehicles and there are vehicles at a distance that will result in a low score, the average value will be low, so the average value may be calculated using only nearby candidates that will give a high score.Alternatively, the average value may be calculated by weighting the score of each object, such as by adding a value according to distance or size to correct the low score, and then the weighted average value may be calculated and compared with a threshold value.

[0070] In addition to information obtained from the output results of the same frame, information about the same object contained in past frames or an object that has been detected with a high score in the recognition results may be corrected to a higher score so that it can be recognized even if the score drops in subsequent frames due to noise or being in the shadow of another object.

[0071] The detection results from image data only contain width and height information, but the detection results from point cloud data contain depth information, so three-dimensional information is used.However, when the target object is directly in front, depth information cannot be obtained, so it is also possible to make a judgment using only width and height information, as with images.

[0072] The threshold may then be determined by combining the corrected score value, the various pieces of information added to the detection result, and the corrected score value.

[0073] So far, an example has been described in which the score of the detection result is used to determine whether to output it as a recognition result, but the recognition result may be output by adding information not included in the detection result based on other parameters and determining whether to output it as a recognition result. For example, if the detection result does not include a type or if a type collectively referred to as "vehicle" is defined, information classifying the detected object into "truck," "passenger car," "light car," etc., using differences in size may be added to the recognition result.

[0074] Furthermore, although the detection result does not include information on the vehicle's direction of travel (whether the side facing the sensor is the front or rear), the output processing unit 106 may add information on the object's direction of travel and speed information, including whether it is stopped, to the recognition result by using map information to identify the lane in which it is traveling, or by calculating the speed from the difference in position in the previous frame.

[0075] <Example 2> Next, a description will be given of a second embodiment of the present invention. In the second embodiment, differences from the first embodiment will be mainly described, and descriptions of the same configurations and processes as the first embodiment will be omitted.

[0076] Systems that use AI to recognize objects have a black box internal process, and may not output the expected results, so they may use other parameters to correct or add to the detection results and output the recognition results. If the system uses AI, such corrections and additions may be made to the recognition results, even in systems that do not recognize objects.

[0077] Furthermore, even when using rule-based object recognition rather than AI recognition, the output from a program that cannot be modified may not produce the expected results. Therefore, other information may be used to correct or process the output results before outputting the final results.

[0078] As described above, according to an embodiment of the present invention, a recognition system for recognizing objects includes a computing device that executes a program and a storage device accessible by the computing device, and includes an object detection processing unit 103 that uses a detection model to extract an area where an object to be extracted may exist from input image data or point cloud data, and outputs the detection result together with a score indicating the likelihood of the detection result, a recognition determination processing unit 104 that generates a recognition result based on the result of evaluating the detection result score using a predetermined threshold, and an output processing unit 106 that outputs the recognition result, where the recognition determination processing unit 104 changes the predetermined threshold depending on a parameter related to the detection result (e.g., the distance to the object), thereby increasing the chances that an object that was previously difficult to recognize becomes recognizable, improving the recognition rate, and outputting an appropriate recognition result. Therefore, if objects such as vehicles ahead of the vehicle or fallen objects can be detected at a long distance, the driver has more time to operate the vehicle and can safely control the vehicle with ease.

[0079] The present invention is not limited to the above-described embodiments, but includes various modifications and equivalent configurations within the spirit and scope of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to configurations including all of the described configurations. Furthermore, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment may be added, deleted, or replaced with other configurations.

[0080] Furthermore, the aforementioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by having a processor interpret and execute a program that realizes each function.

[0081] Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, a hard disk, or an SSD (Solid State Drive), or in a recording medium such as an IC card, an SD card, or a DVD.

[0082] In addition, the control lines and information lines shown are those that are considered necessary for explanation, and do not necessarily represent all the control lines and information lines that are necessary for implementation. In reality, it can be assumed that almost all components are interconnected.

Claims

1. A recognition system for recognizing an object, comprising: A computing device that executes a program and a storage device that can be accessed by the computing device, an object detection processing unit configured to use a detection model to extract an area where an object to be extracted may exist from input image data or point cloud data, and to output the detection result together with a score representing the likelihood of the detection result; a recognition determination processing unit configured to generate a recognition result based on a result of evaluating the score of the detection result using a predetermined threshold; the arithmetic device has an output processing unit that outputs the recognition result, The recognition system is characterized in that the recognition determination processing unit changes the predetermined threshold value according to the distance to the detected object using a quadratic function that has little change in the vicinity.

2. 2. The recognition system of claim 1, A recognition system characterized in that the quadratic function represents whether the area is drivable by a vehicle.

3. 3. A recognition system according to claim 1 or 2, The output processing unit adds image data, point cloud data, or information acquired by another sensor to the recognition result and outputs the result.

4. 3. A recognition system according to claim 1 or 2, A recognition system characterized in that the detection model is a trained neural network model or a rule-based model.

5. An object recognition method executed by a recognition system that recognizes an object, comprising: the recognition system includes a computing device that executes a program and a memory accessible by the computing device; The object recognition method includes: the computing device uses a detection model to extract an area where an object to be extracted may exist from the input image data or point cloud data, and outputs the detection result together with a score representing the likelihood of the detection result; the computing device generates a recognition result based on a result of evaluating the score of the detection result using a predetermined threshold; The computing device outputs the recognition result, An object recognition method, characterized in that the predetermined threshold is changed according to the distance to the detected object using a quadratic function that has little change in the vicinity.

Images