Recognition and processing method of image detection based on unity model bounding box

Abstract

The invention discloses a recognition and processing method for a model bounding box detection image based on Unity, and belongs to the technical field of image recognition. The method comprises: importing a model file in Unity, creating a camera object and a model object to be detected, and marking them as a target model; performing bounding box detection on the target model based on the model bounding box detection function of Unity, obtaining the position, size and direction information of the target model, and generating a target bounding box; performing real-time image acquisition on the target bounding box through the camera function of Unity, and performing corresponding preprocessing to obtain a detection target image; establishing an image recognition model based on a neural network, optimizing and analyzing the image recognition model, and determining a target recognition model; recognizing the detection target image through the target recognition model, and outputting a recognition result; and interacting the target model with the camera according to the recognition result.

Description

Technical Field

[0001] The invention belongs to the technical field of image recognition, and in particular is a recognition processing method for a model bounding box detection image based on Unity. Background Art

[0002] As an important branch of computer vision, image recognition technology has made significant progress in recent years and is widely used in security monitoring, autonomous driving, smart medical care, robot navigation and other fields. However, in practical applications, image recognition technology still faces many challenges, such as large differences in images under different lighting conditions, which affects recognition accuracy; recognition difficulty increases when objects are partially blocked or photographed from different angles; target recognition in complex backgrounds is easily disturbed, reducing recognition efficiency; and in application scenarios with high real-time requirements, such as games and virtual reality, image information needs to be quickly responded to and processed.

[0003] At present, image recognition methods mainly include traditional methods based on feature extraction and modern methods based on deep learning. Although deep learning has achieved great success in the field of image recognition, it still has certain limitations.

[0004] Therefore, in order to optimize image recognition, the present invention provides a recognition and processing method for detecting images using a model bounding box based on Unity. Summary of the invention

[0005] In order to solve the problems existing in the above scheme, the present invention provides a recognition and processing method for detecting an image of a model bounding box based on Unity.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A recognition and processing method for detecting an image with a bounding box of a model based on Unity, the method comprising:

[0008] Step 1: Import the model file into Unity, create a camera object and the model object to be detected, and mark it as the target model;

[0009] Step 2: Perform bounding box detection on the target model based on Unity's model bounding box detection function, obtain the position, size and direction information of the target model, and generate the target bounding box;

[0010] Step 3: Use the Unity camera function to collect real-time images of the target bounding box and perform corresponding preprocessing to obtain the detected target image;

[0011] Step 4: Establish an image recognition model based on a neural network, optimize and analyze the image recognition model, and determine a target recognition model;

[0012] Furthermore, the method for optimizing and analyzing the image recognition model includes:

[0013] Set various influencing factors and influencing conditions corresponding to the influencing factors; perform scenario simulation according to the influencing conditions corresponding to the influencing factors to form various simulation scenarios;

[0014] Set material data, verify the image recognition model through the material data in each simulation scene, and obtain the recognition accuracy and recognition efficiency of the image recognition model in each simulation scene;

[0015] Obtain benchmark accuracy and benchmark efficiency;

[0016] According to the formula Calculate the corresponding scene evaluation value;

[0017] Where: PA is the scene evaluation value; exp() is an exponential function with the natural constant e as the base; ZQ0 is the benchmark accuracy; ZQ1 is the recognition accuracy of the corresponding simulation scene; XQ0 is the benchmark efficiency; XQ1 is the recognition efficiency of the corresponding simulation scene; b1 and b2 are both proportional coefficients, with a value range of 0 <b1≤1,0<b2≤1;

[0018] When the scenario evaluation value is not greater than the threshold value X1, the evaluation is adapted to the corresponding simulation scenario;

[0019] When the scenario evaluation value is greater than the threshold X1, the evaluation is not suitable for the corresponding simulation scenario;

[0020] Each unsuitable simulation scene is marked as an abnormal scene, and the image recognition model is optimized and adjusted according to each abnormal scene to obtain a target recognition model.

[0021] Step 5: Identify the detection target image through the target recognition model and output the recognition result;

[0022] Step six: interacting the target model with the camera according to the recognition result.

[0023] Furthermore, it also includes step seven: performing real-time monitoring and analysis during the identification process to obtain monitoring and analysis results.

[0024] Furthermore, the method of performing real-time monitoring and analysis during the identification process includes:

[0025] Establishing a dynamic verification scheme, and randomly outputting a reference target based on the dynamic verification scheme;

[0026] Perform packaging box inspection and verification based on the reference target to obtain verification results, which include verification abnormality and verification normality;

[0027] When the verification result is abnormal, corresponding warning processing is carried out;

[0028] When the verification result is that the verification is normal, an identification evaluation is performed to obtain an identification evaluation result, wherein the identification evaluation result includes normal identification and abnormal identification; and corresponding processing is performed according to the identification evaluation result.

[0029] Furthermore, the method for establishing a dynamic verification scheme includes:

[0030] Identify the target model in real time, determine the target range according to the target model, define the retrieval standard according to the target range and the target recognition model, the retrieval standard is that the retrieved target meets the target range and conforms to the recognition of the target recognition model; obtain each reference target according to the retrieval standard;

[0031] Identify each reference target through the target recognition model to obtain recognition results of each reference target, determine the detection target corresponding to each reference target according to the recognition results; generate a reference standard target corresponding to the reference target according to the detection target and the recognition results;

[0032] Establish a reference library according to each reference target and reference standard target;

[0033] Setting a random unit, the random unit is used to randomly output a reference target from the reference library within a preset time interval;

[0034] The reference targets in the reference library are randomly outputted through the random unit.

[0035] Furthermore, the method for performing packaging box detection and verification based on the reference target includes:

[0036] Based on Unity's bounding box detection function, perform bounding box detection on the reference target and generate the target bounding box;

[0037] Obtain a reference standard target corresponding to the reference target, identify the target bounding box in the reference standard target, and mark it as a standard bounding box;

[0038] Calculate the corresponding difference value based on the standard bounding box and the target bounding box;

[0039] Mark the corresponding detection target in the target bounding box according to the detection target in the reference standard target;

[0040] Establish an encirclement judgment model, the expression of the encirclement judgment model is:

[0041]

[0042] Where: s is the input data, the input data is the target bounding box marking the detection target; the output data is the abnormal judgment value DG(s);

[0043] Analyze the target bounding box by using the bounding judgment model to obtain a corresponding abnormality judgment value;

[0044] Calculate the corresponding encirclement verification value according to the formula PB = DG (s) × (CY + 0.1);

[0045] Where: PB is the bracket verification value; DG(s) is the abnormal judgment value; CY is the difference value;

[0046] The verification result is determined based on the bracketing verification values.

[0047] Further, the method of performing identification assessment includes:

[0048] Identify the target bounding box corresponding to the reference target, collect the corresponding detection target image, recognize the detection target image through the target recognition model, and obtain the recognition result;

[0049] Obtain a reference standard target corresponding to the reference target, identify a recognition result corresponding to the reference standard target, mark it as a recognition standard result, compare the recognition standard result with the recognition result, and determine a recognition evaluation result.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] Improve recognition accuracy: By combining model bounding box detection technology, the position and posture of the target can be determined more accurately, thereby improving the accuracy of image recognition.

[0052] Real-time and stability: It can realize real-time image recognition on the Unity platform, and has strong stability and robustness, which is suitable for application requirements in various complex scenarios.

[0053] Easy to integrate and apply: As a Unity-based solution, it is easy to integrate into existing Unity applications and can work seamlessly with other Unity functions (such as the physics engine, animation system, etc.), providing a convenient development and application experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0056] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] like Figure 1 As shown, a recognition and processing method for detecting an image based on a model bounding box of Unity includes:

[0058] Step 1: Import the model file into Unity, create a camera object and the model object to be detected, and mark it as the target model;

[0059] Step 2: Perform bounding box detection on the model based on Unity's model bounding box detection function, obtain the position, size and direction information of the target model, generate a bounding box, and mark it as the target bounding box.

[0060] Exemplary:

[0061] Use target detection algorithms (such as YOLO, SSD, Faster R-CNN, etc.) to find and locate target objects in the preprocessed data; these algorithms can output the location information (such as bounding box coordinates) and category information (such as vehicles, pedestrians, etc.) of the detected targets;

[0062] Based on the results of target detection, a bounding box is generated to more tightly surround the target object. The bounding box can be axis-aligned (such as AABB) or directional (such as OBB), depending on the detection requirements and the shape characteristics of the target object.

[0063] The resulting bounding boxes are further optimized and adjusted to improve their accuracy and tightness. This can include the application of algorithms such as non-maximum suppression (NMS) to reduce duplicate detections and false positives.

[0064] Step 3: Use the camera function of Unity to collect real-time images of the target bounding box and perform preprocessing operations, such as color space conversion and image enhancement, to obtain the detected target image;

[0065] Step 4: Establish an image recognition model based on a neural network such as a CNN network or a DNN network, optimize and analyze the image recognition model, and determine the target recognition model;

[0066] For example, image recognition algorithms based on deep learning are combined with convolutional neural networks and other technologies to establish image recognition models.

[0067] In one embodiment, the image recognition model may not be optimized and analyzed, and may be directly marked as a target recognition model.

[0068] In one embodiment, since Unity is a widely used game development engine, it does not directly determine the recognition ability of the image recognition model. However, when Unity integrates and uses these models, it may be affected by some factors, which in turn indirectly affects the effect of image recognition. For example, when Unity runs on mobile devices or low-end hardware, it may be limited by the processor and memory. This may cause the image recognition model to run slower, increase response delays, and affect real-time performance. When multiple complex tasks (such as image recognition, physical simulation, animation, etc.) are run simultaneously in the Unity scene, the frame rate may drop, which in turn affects the smoothness and accuracy of image recognition. Unity supports multi-platform releases, but the performance and hardware characteristics of different platforms (such as Android, iOS, PC, etc.) are different. This may cause the image recognition model to perform inconsistently on different platforms. There are many factors that affect the recognition of the image recognition model, so corresponding adaptive optimization is required.

[0069] Methods for optimizing and analyzing image recognition models include:

[0070] Set material data based on subsequent image recognition requirements. The material data can be regarded as a validation set for the image recognition model, or it can be set based on the validation set or training set.

[0071] According to the historical impact record of Unity on the image recognition model, each influencing factor and the corresponding influencing condition of each influencing factor are set; the influencing condition is the possible impact of the corresponding influencing factor on the recognition of the image recognition model in this case according to the statistics of the historical impact record;

[0072] That is, various historical record data of the image recognition model under the unity background are summarized, and various influencing factors that have an impact on the recognition of the image recognition model are screened out from the historical record data, and the influencing conditions corresponding to the influencing factors are determined according to the corresponding historical record data, which are used for subsequent scene simulation in combination with the influencing factors and influencing conditions.

[0073] Perform scene simulation according to the influencing conditions corresponding to each influencing factor, that is, use Unity to simulate each influencing condition to form each simulation scene;

[0074] The image recognition model is verified by using the material data in each simulation scenario to obtain the recognition accuracy and recognition efficiency of the image recognition model in each simulation scenario. The recognition efficiency is expressed by taking the average value;

[0075] The preset recognition accuracy and recognition efficiency of the image recognition model are marked as the benchmark accuracy and benchmark efficiency respectively; the preset recognition accuracy is the recognition accuracy that the image recognition model needs to achieve, that is, it is set according to the actual recognition accuracy requirements, and can also be regarded as the expected accuracy, such as 98%;

[0076] According to the formula Calculate the corresponding scene evaluation value;

[0077] Where: PA is the scene evaluation value; exp() is an exponential function with the natural constant e as the base; ZQ0 is the benchmark accuracy; ZQ1 is the recognition accuracy of the corresponding simulation scene, and is the recognition accuracy calculated based on the simulation results; XQ0 is the benchmark efficiency; XQ1 is the recognition efficiency of the corresponding simulation scene; b1 and b2 are both proportional coefficients, with a value range of 0 <b1≤1,0<b2≤1;

[0078] When the scenario evaluation value is not greater than the threshold value X1, the evaluation is adapted to the corresponding simulation scenario;

[0079] When the scenario evaluation value is greater than the threshold X1, the evaluation is not suitable for the corresponding simulation scenario;

[0080] Each unsuitable simulation scene is marked as an abnormal scene, and the image recognition model is optimized and adjusted according to each abnormal scene to obtain a target recognition model.

[0081] The image recognition model is optimized and adjusted according to each abnormal scenario, that is, the existing technology is used to overcome the abnormal scenarios that cannot be applied, such as adjusting the program, adjusting the parameters, adjusting the training, etc., so that it can overcome the abnormal scenarios; or if the user determines that the abnormal scenarios do not need to be overcome, there is no need to perform overcoming optimization.

[0082] The threshold X1 is set by a technician in this field according to actual conditions or obtained by simulating a large amount of data; or it is set by the user. For example, the user sets the allowable deviation that he or she expects to achieve, that is, within the allowable deviation, the allowable deviation is (ZQ0-ZQ1) / ZQ0, (XQ0-XQ1) / XQ0, that is, the allowable deviation that can be had, and then substituted into the formula to calculate the corresponding scenario evaluation value, which is regarded as the threshold X1. If it exceeds its value, it is regarded as exceeding the allowable deviation, and the evaluation is not suitable for the corresponding simulation scenario.

[0083] After determining each abnormal scenario, the user decides whether to perform optimization. If optimization is performed, optimization adjustments are made according to the corresponding abnormal scenario to overcome the corresponding problem.

[0084] Step 5: Identify the detection target image through the target recognition model, identify the position of the target model in the scene, and output the recognition result.

[0085] Step 6: Based on the recognition results, realize the interaction between the target model and the camera, such as marking the recognized position on the target model, playing animation, etc.

[0086] In one embodiment, because the recognition is based on Unity, the actual recognition result will be affected by Unity. For example, multiple tasks running simultaneously in Unity may lead to performance bottlenecks, making it impossible for the image recognition model to process images in real time; bounding box detection is abnormal, resulting in bounding box generation errors; improper resource management in Unity (such as memory leaks, poor resource management, etc.) may lead to various abnormal situations such as program crashes or slow operation; but users cannot monitor Unity, resulting in the inability to timely identify problems when Unity has abnormal situations; in particular, it is difficult to judge the recognition results manually in real time; therefore, in order to ensure true and effective recognition during use, corresponding process monitoring is required, which also includes step seven: real-time monitoring and analysis during the recognition process to obtain monitoring and analysis results. The process is as follows:

[0087] Establish a dynamic verification scheme, and randomly output reference targets based on the dynamic verification scheme;

[0088] Perform packaging box inspection and verification based on the reference target to obtain verification results, which include verification abnormality and verification normality;

[0089] When the verification result is abnormal, corresponding warning processing is carried out;

[0090] When the verification result is normal, an identification evaluation is performed to obtain an identification evaluation result, which includes normal identification and abnormal identification. Corresponding processing is performed according to the identification evaluation result, such as early warning and other processing methods.

[0091] The method for establishing a dynamic verification scheme includes:

[0092] Identify the target model in step 1 in real time, determine the corresponding target range such as type and field according to the target model, search according to the target range under the premise of good network status, and obtain the reference target within the target range and in line with the recognition requirements of the target recognition model. The reference target is the model target that may appear in the recognition process, such as vehicle recognition, and the reference target is each possible vehicle, which can be recognized by the target recognition model under normal conditions; that is, dynamically update according to the target model;

[0093] Each reference target is identified through the target recognition model to obtain the recognition result, and the detection target corresponding to each reference target is determined according to the recognition result. The reference standard target corresponding to the reference target is generated according to the generation method of the target bounding box, the detection target and the recognition result; because there is no timeliness requirement, the reference target can be directly identified at any time through the target recognition model to determine the detection target, such as vehicles, personnel, products, etc.; then the correct target bounding box is determined according to the generation method of the detection target and the target bounding box, and the reference standard target is summarized; a reference library is established according to each reference target and the reference standard target, and the reference library can be updated if conditions permit, so as to avoid always applying the same reference targets;

[0094] Setting a random unit, the random unit is used to randomly output a corresponding reference target from a reference library within a preset time interval;

[0095] The reference targets in the reference library are randomly outputted through the random unit.

[0096] In one embodiment, a method for performing packaging box detection and verification based on a reference target includes:

[0097] Based on Unity's bounding box detection function, perform bounding box detection on the reference target and generate the target bounding box;

[0098] Obtain a reference standard target corresponding to the reference target, identify the target bounding box in the reference standard target, and mark it as a standard bounding box;

[0099] Compare the standard bounding box with the target bounding box, calculate the corresponding difference value, calculate the difference rate based on the position of the bounding box and other parameters, and mark it as the difference value;

[0100] The corresponding difference evaluation items are mainly determined based on the factors that have an impact on the subsequent identification based on the bounding box. For example, the position difference and the bounding box shape difference may lead to anomalies, such as failure to completely include the target, normal comparison offset, non-optimal packaging box, etc., that is, the difference rate is mainly calculated based on the position and shape of the two packaging boxes, and the corresponding proportional coefficient is preset for calculation; it can also be calculated based on various existing difference rate calculation methods combined with various difference evaluation items.

[0101] And according to the detection target in the standard bounding box, the target bounding box is judged as abnormal, and the corresponding abnormal judgment value is obtained, that is, whether the target bounding box completely includes the detection target. If it completely includes the detection target, it meets the abnormal judgment requirement. Completely including the detection target means that the entire detection target is in the target bounding box; the abnormal judgment value is 1, otherwise it is 0; specifically, an encirclement judgment model can be established, and the target bounding box can be judged as abnormal through the encirclement judgment model. The expression of the encirclement judgment model is Where: s is the input data, the input data is the target bounding box marking the detection target; the output data is the abnormal judgment value DG(s); the corresponding abnormal judgment value is obtained by analyzing the bounding box judgment model;

[0102] Calculate the corresponding encirclement verification value according to the formula PB = DG (s) × (CY + 0.1);

[0103] Where: PB is the bracket verification value; DG(s) is the abnormal judgment value; CY is the difference value;

[0104] When the encirclement verification value is greater than a preset value or equal to 0, the verification is determined to be abnormal, otherwise the verification is determined to be normal.

[0105] In one embodiment, the packaging box detection and verification is performed based on a reference target, and verification is performed as a separate task to avoid affecting the ongoing identification process.

[0106] Methods for conducting identification assessments include:

[0107] Identify the target bounding box corresponding to the reference target, and then analyze the above embodiment; collect the corresponding detection target image, and recognize the detection target image through the target recognition model to obtain the recognition result;

[0108] Obtain a reference standard target corresponding to the reference target, identify a recognition result corresponding to the reference standard target, mark it as a recognition standard result, compare the recognition standard result with the recognition result, and determine a recognition evaluation result.

[0109] The recognition standard result and the recognition result are the corresponding image recognition results respectively. The two are compared, that is, whether the two image recognition results are the same. If they are the same, the recognition is normal, otherwise the recognition is abnormal.

[0110] The above formulas are all calculated by removing dimensions and taking numerical values. The formula is a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data.

[0111] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A recognition and processing method for detecting an image with a bounding box based on a unity model, characterized in that: Methods include: Import the model file into Unity, create a camera object and the model object that needs to be detected, and mark it as the target model; Based on Unity's model bounding box detection function, perform bounding box detection on the target model, obtain the position, size and direction information of the target model, and generate the target bounding box; The target bounding box is captured in real time through the Unity camera function, and the corresponding preprocessing is performed to obtain the detected target image; Establishing an image recognition model based on a neural network, optimizing and analyzing the image recognition model, and determining a target recognition model; Identify the detection target image through the target recognition model and output the recognition result; interacting the target model with a camera according to the recognition result; Methods for optimizing and analyzing image recognition models include: Set various influencing factors and influencing conditions corresponding to the influencing factors; perform scenario simulation according to the influencing conditions corresponding to the influencing factors to form various simulation scenarios; Set material data, verify the image recognition model through the material data in each simulation scene, and obtain the recognition accuracy and recognition efficiency of the image recognition model in each simulation scene; Obtain benchmark accuracy and benchmark efficiency; According to the formula Calculate the corresponding scene evaluation value; Where: PA is the scene evaluation value; exp() is an exponential function with the natural constant e as the base; ZQ0 is the benchmark accuracy; ZQ1 is the recognition accuracy of the corresponding simulation scene; XQ0 is the benchmark efficiency; XQ1 is the recognition efficiency of the corresponding simulation scene; b1 and b2 are both proportional coefficients, with a value range of 0 <b1≤1,0<b2≤1; When the scenario evaluation value is not greater than the threshold value X1, the evaluation is adapted to the corresponding simulation scenario; When the scenario evaluation value is greater than the threshold X1, the evaluation is not suitable for the corresponding simulation scenario; Mark each unsuitable simulation scene as an abnormal scene, optimize and adjust the image recognition model according to each abnormal scene, and obtain a target recognition model; Conduct real-time monitoring and analysis during the identification process to obtain monitoring and analysis results; Methods for real-time monitoring and analysis during the identification process include: Establishing a dynamic verification scheme, and randomly outputting a reference target based on the dynamic verification scheme; Perform packaging box inspection and verification based on the reference target to obtain verification results, which include verification abnormality and verification normality; When the verification result is abnormal, corresponding warning processing is carried out; When the verification result is that the verification is normal, an identification evaluation is performed to obtain an identification evaluation result, wherein the identification evaluation result includes normal identification and abnormal identification; and corresponding processing is performed according to the identification evaluation result.

2. The recognition and processing method of the model bounding box detection image based on Unity according to claim 1 is characterized in that: The method for establishing a dynamic verification scheme includes: Identify the target model in real time, determine the target range according to the target model, define the retrieval criteria according to the target range and the target identification model; obtain each reference target according to the retrieval criteria; Identify each reference target through the target recognition model to obtain recognition results of each reference target, determine the detection target corresponding to each reference target according to the recognition results; generate a reference standard target corresponding to the reference target according to the detection target and the recognition results; Establish a reference library according to each reference target and reference standard target; Setting a random unit, the random unit is used to randomly output a reference target from the reference library within a preset time interval; The reference targets in the reference library are randomly outputted through the random unit.

3. The recognition and processing method of the model bounding box detection image based on Unity according to claim 2 is characterized in that: Methods for packaging box inspection and verification based on reference targets include: Based on Unity's bounding box detection function, perform bounding box detection on the reference target and generate the target bounding box; Obtain a reference standard target corresponding to the reference target, identify the target bounding box in the reference standard target, and mark it as a standard bounding box; Calculate the corresponding difference value based on the standard bounding box and the target bounding box; Mark the corresponding detection target in the target bounding box according to the detection target in the reference standard target; Establish an encirclement judgment model, the expression of the encirclement judgment model is: Where: s is the input data, the input data is the target bounding box marking the detection target; the output data is the abnormal judgment value DG(s); Analyze the target bounding box by using the bounding judgment model to obtain a corresponding abnormality judgment value; Calculate the corresponding encirclement verification value according to the formula PB = DG (s) × (CY + 0.1); Where: PB is the bracket verification value; DG(s) is the abnormal judgment value; CY is the difference value; The verification result is determined based on the bracketing verification values.

4. The recognition and processing method of the model bounding box detection image based on Unity according to claim 3 is characterized in that: Methods for conducting identification assessments include: Identify the target bounding box corresponding to the reference target, collect the corresponding detection target image, recognize the detection target image through the target recognition model, and obtain the recognition result; Obtain a reference standard target corresponding to the reference target, identify a recognition result corresponding to the reference standard target, mark it as a recognition standard result, compare the recognition standard result with the recognition result, and determine a recognition evaluation result.

Images