CN110968718A - Target detection model negative sample mining method and device and electronic equipment - Google Patents
Target detection model negative sample mining method and device and electronic equipment Download PDFInfo
- Publication number
- CN110968718A CN110968718A CN201911136309.2A CN201911136309A CN110968718A CN 110968718 A CN110968718 A CN 110968718A CN 201911136309 A CN201911136309 A CN 201911136309A CN 110968718 A CN110968718 A CN 110968718A
- Authority
- CN
- China
- Prior art keywords
- target
- picture
- detection
- negative sample
- target picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Quality & Reliability (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a target detection model negative sample mining method and device and electronic equipment, relates to the field of machine learning, and can be used for automatic driving. The specific implementation scheme is as follows: carrying out target detection on a target picture to be detected by using the generated target detection model to obtain a detection result corresponding to the target picture; judging whether the detection result meets a preset condition or not; and if so, determining that the target picture is a negative sample picture. Through the scheme of this application, can realize the automatic excavation of negative sample picture, the process of excavating the negative sample picture need not artifical the participation, the cost of using manpower sparingly, compare in the mode of artifical excavation, the speed of excavating the negative sample picture automatically is very fast, and is consuming time less, improves and excavates efficiency, for reliability and the accuracy that improve the target detection model provide the condition, solve the artifical technical problem who selects the negative sample and consume time length, inefficiency among the prior art.
Description
Technical Field
The application relates to the technical field of computer technology and machine learning, in particular to a target detection model negative sample mining method, device and electronic equipment, which can be used for automatic driving.
Background
In order to reduce the probability of missing detection and false detection of the target detection model in the target detection process, a negative sample is usually added to train the target detection model when the target detection model is trained.
At present, negative samples are mainly obtained in a manual selection mode, and images with poor model detection effects are manually selected from a standard data set or a private data set. However, this negative sample acquisition method requires a lot of labor cost, and is time-consuming and inefficient.
Disclosure of Invention
The application provides a target detection model negative sample mining method and device and electronic equipment, and aims to solve the technical problems that manual negative sample selection in the prior art is long in time consumption and low in efficiency.
An embodiment of a first aspect of the present application provides a target detection model negative sample mining method, including:
carrying out target detection on a target picture to be detected by using the generated target detection model to obtain a detection result corresponding to the target picture;
judging whether the detection result meets a preset condition or not, wherein the preset condition comprises at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold;
and if so, determining that the target picture is a negative sample picture.
According to the target detection model negative sample mining method, the generated target detection model is utilized, the target picture to be detected is subjected to target detection, the detection result corresponding to the target picture is obtained, whether the detection result meets the preset condition or not is judged, and when the detection result meets the preset condition, the target picture is determined to be the negative sample picture. Therefore, whether the target picture is the negative sample picture or not is judged by judging whether the detection result meets the preset condition or not, when the detection result meets the preset condition, the target picture is determined to be the negative sample picture, automatic excavation of the negative sample picture is achieved, manual participation is not needed in the process of excavating the negative sample picture, labor cost is saved, compared with a manual excavation mode, the speed of automatically excavating the negative sample picture is high, time consumption is low, excavation efficiency is improved, and conditions are provided for improving reliability and accuracy of a target detection model.
In a possible implementation manner of the embodiment of the present application, after determining whether the detection result meets a preset condition, the method further includes:
if not, judging whether the detection result contains a target object, wherein the target object is an object detected by the target detection model in N frames of continuous pictures before the target picture is detected, and N is an integer which is greater than 0 and smaller than a fourth threshold value;
and if not, determining that the N frames of continuous pictures before the target picture are negative sample pictures.
Therefore, when the detection result does not meet the preset condition, whether the target object is contained in the detection result is further judged, and when the target object is not contained, the N frames of continuous pictures before the target picture are determined as the negative sample pictures, so that the mining of the false detection samples is realized, and the mining accuracy of the negative sample pictures is improved.
In a possible implementation manner of the embodiment of the present application, after determining whether the detection result meets a preset condition, the method further includes:
if the current image is not met, identifying the target image by using a preset background identification model so as to determine a foreground area and a background area contained in the target image;
judging whether the background region contains a detection object or not according to a detection result corresponding to the target picture;
and if so, determining that the target picture is a negative sample picture.
Therefore, when the detection result does not meet the preset condition, the negative sample picture is mined by further combining the recognition result of the background recognition model, and the target picture containing the detection object in the background area is determined as the negative sample picture, so that the mining of the false detection sample is realized, and the coverage rate and the accuracy of the mining of the negative sample picture are improved.
In a possible implementation manner of the embodiment of the present application, after determining the foreground region and the background region included in the target picture, the method further includes:
judging whether the foreground region contains a detection object or not according to a detection result corresponding to the target picture;
if not, determining that the target picture is a negative sample picture.
Therefore, when the detection result does not meet the preset condition, the negative sample picture is mined by further combining the identification result of the background identification model, and the target picture which does not contain the detection object in the foreground area is determined as the negative sample picture, so that the mining of the sample picture containing the detection object with weak detection capability is realized, and the coverage rate and the accuracy of the mining of the negative sample picture are improved.
In a possible implementation manner of the embodiment of the present application, after determining that the target picture is a negative sample picture, the method further includes:
and performing correction training on the generated target detection model by using the determined negative sample picture.
Therefore, the target detection model is corrected and trained by using the determined negative sample picture, and the reliability and the accuracy of the target detection model are improved.
The embodiment of the second aspect of the present application provides a target detection model negative sample mining device, including:
the acquisition module is used for carrying out target detection on a target picture to be detected by utilizing the generated target detection model and acquiring a detection result corresponding to the target picture;
the first judging module is configured to judge whether the detection result meets a preset condition, where the preset condition includes at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold;
and the determining module is used for determining that the target picture is a negative sample picture when the detection result meets a preset condition.
According to the target detection model negative sample mining device, the generated target detection model is utilized, the target picture to be detected is subjected to target detection, the detection result corresponding to the target picture is obtained, whether the detection result meets the preset condition or not is judged, and when the detection result meets the preset condition, the target picture is determined to be the negative sample picture. Therefore, whether the target picture is the negative sample picture or not is judged by judging whether the detection result meets the preset condition or not, when the detection result meets the preset condition, the target picture is determined to be the negative sample picture, automatic excavation of the negative sample picture is achieved, manual participation is not needed in the process of excavating the negative sample picture, labor cost is saved, compared with a manual excavation mode, the speed of automatically excavating the negative sample picture is high, time consumption is low, excavation efficiency is improved, and conditions are provided for improving reliability and accuracy of a target detection model.
In a possible implementation manner of the embodiment of the present application, the apparatus further includes:
a second determining module, configured to determine whether a target object is included in the detection result when the detection result does not meet a preset condition, where the target object is an object detected by the target detection model in N consecutive pictures before the target picture is detected, and N is an integer greater than 0 and smaller than a fourth threshold;
the determining module is further configured to:
and if the detection result does not contain the target object, determining that the N frames of continuous pictures before the target picture are negative sample pictures.
In a possible implementation manner of the embodiment of the present application, the apparatus further includes:
the processing module is used for identifying the target picture by using a preset background identification model when the detection result does not meet a preset condition so as to determine a foreground area and a background area contained in the target picture;
the third judging module is used for judging whether the background area contains a detection object or not according to the detection result corresponding to the target picture;
the determining module is further configured to:
and if the background region contains the detection object, determining that the target picture is a negative sample picture.
In a possible implementation manner of the embodiment of the present application, the third determining module is further configured to:
judging whether the foreground region contains a detection object or not according to a detection result corresponding to the target picture;
the determining module is further configured to:
and if the foreground region does not contain the detection object, determining that the target picture is a negative sample picture.
In a possible implementation manner of the embodiment of the present application, the apparatus further includes:
and the correction module is used for performing correction training on the generated target detection model by using the determined negative sample picture.
An embodiment of a third aspect of the present application provides an electronic device, including:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of negative sample mining of an object detection model as described in the foregoing embodiments of the first aspect.
A fourth aspect of the present application provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the target detection model negative sample mining method described in the foregoing first aspect.
One embodiment in the above application has the following advantages or benefits:
the target detection method comprises the steps of carrying out target detection on a target picture to be detected by utilizing a generated target detection model, obtaining a detection result corresponding to the target picture, judging whether the detection result meets a preset condition or not, and determining the target picture as a negative sample picture when the detection result meets the preset condition. Therefore, whether the target picture is the negative sample picture or not is judged by judging whether the detection result meets the preset condition or not, when the detection result meets the preset condition, the target picture is determined to be the negative sample picture, automatic excavation of the negative sample picture is achieved, manual participation is not needed in the process of excavating the negative sample picture, labor cost is saved, compared with a manual excavation mode, the speed of automatically excavating the negative sample picture is high, time consumption is low, excavation efficiency is improved, and conditions are provided for improving reliability and accuracy of a target detection model. The technical means of judging whether the detection result of the target detection model on the target picture meets the preset condition or not and determining the target picture as the negative sample picture when the preset condition is met is adopted, and manual participation is not needed, so that the technical problems of long time consumption and low efficiency in manual selection of the negative sample in the prior art are solved, and the technical effects of automatically mining the negative sample picture and improving mining efficiency are achieved.
Other effects of the above-described alternative will be described below with reference to specific embodiments.
Drawings
The drawings are included to provide a better understanding of the present solution and are not intended to limit the present application. Wherein:
FIG. 1 is a schematic flow chart diagram of a target detection model negative sample mining method according to a first embodiment of the present application;
FIG. 2 is a schematic flow chart diagram of a target detection model negative sample mining method according to an embodiment of the second aspect of the present application;
FIG. 3 is a schematic flow chart diagram of a target detection model negative sample mining method according to an embodiment of the third aspect of the present application;
fig. 4 is a schematic structural diagram of a target detection model negative sample mining device according to a fourth embodiment of the present application;
fig. 5 is a schematic structural diagram of a target detection model negative sample mining device according to a fifth embodiment of the present application;
fig. 6 is a schematic structural diagram of a target detection model negative sample mining device according to a sixth embodiment of the present application;
fig. 7 is a schematic structural diagram of a target detection model negative sample mining device according to a seventh embodiment of the present application;
fig. 8 is a block diagram of an electronic device for implementing a target detection model negative sample mining method according to an embodiment of the present application.
Detailed Description
The following description of the exemplary embodiments of the present application, taken in conjunction with the accompanying drawings, includes various details of the embodiments of the application for the understanding of the same, which are to be considered exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The target detection model negative sample mining method, device and electronic equipment of the present application are described below with reference to the accompanying drawings.
For the target detection model, some negative samples (badcases) are inevitably generated in the use process, and although the occurrence frequency of the negative samples is not high, the use experience of the user is greatly influenced when the negative samples occur, for example, in an unmanned scene, the detected negative samples have a great influence on the path planning of an unmanned vehicle, and even one false detection or missed detection can cause the take over of an automatic driving vehicle.
In order to reduce the probability of false detection and false detection of the target detection model in the detection process, more negative samples are generally collected to correct the target detection model. Currently, the way to collect negative samples generally includes the following:
(1) and (5) manually selecting. The manual selection is a negative sample collection mode with a common target, the images with poor detection effects are obtained by manually selecting the negative samples in a standard data set or a private data set, and the images are extracted and labeled. This approach is labor and time intensive and inefficient.
(2) And collecting the data in a web crawler mode. More data meeting the requirements are acquired in a short time by setting keywords and specific rules, but the data crawled in the mode contains a large amount of noise, and the available badcase still needs to be acquired by manual selection, so that the time is consumed.
(3) And setting up a scene for collection. And (3) building a scene similar to the scene with the badcase, and rapidly and massively collecting data in the built scene. Although the method can solve the problems existing in the first two methods, the cost for setting up the scene is high, the acquisition scene is single, and the acquired badcase is single, so that the generalization capability of the target detection model is insufficient.
In order to solve the problems, the application provides a negative sample mining method of a target detection model, target detection is carried out on a target picture by utilizing the target detection model through presetting negative sample mining conditions, and when a detection result meets the preset conditions, the target picture is determined to be the negative sample picture, so that automatic mining of the negative sample picture is realized, manual participation is not needed in the process of mining the negative sample picture, labor cost is saved, compared with a manual mining mode, the speed of automatically mining the negative sample picture is high, time consumption is low, mining efficiency is improved, and conditions are provided for improving reliability and accuracy of the target detection model.
Specifically, fig. 1 is a schematic flowchart of a target detection model negative sample mining method according to a first embodiment of the present application, and the method may be executed by the target detection model negative sample mining apparatus provided in the present application, and may also be executed by an electronic device, where the electronic device may be a server, or may also be a terminal device such as a desktop computer or a notebook computer, and the present application is not limited thereto. The following explains the present application by taking as an example that the target detection model negative sample mining device proposed in the present application executes the target detection model negative sample mining method of the present application.
As shown in fig. 1, the target detection model negative sample mining method includes the following steps:
and 101, performing target detection on a target picture to be detected by using the generated target detection model, and acquiring a detection result corresponding to the target picture.
The target detection model may be a model obtained through pre-training and used for detecting each detection object in the picture, for example, in an unmanned scene, the target detection model detects a vehicle in the picture. The target picture to be detected can be a picture in a standard data set, a picture in a private data set, a picture acquired by a camera in real time or a certain frame of picture in a video stream acquired in real time, and the like.
In this embodiment, the target image to be detected may be subjected to target detection by using the generated target detection model, and a detection result corresponding to the target image is obtained.
For example, for a picture acquired by a camera at an intersection, a target detection model may be used to perform target detection on a moving object in the picture, or only a vehicle in the picture may be detected.
And 102, judging whether the detection result meets a preset condition or not.
The preset condition refers to a condition that needs to be satisfied when a certain picture is determined as a negative sample picture, and the preset condition may be preset. In an embodiment of the present application, the preset condition includes at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold. The confidence in this embodiment refers to the reliability of the correct recognition of the detection object, and the higher the confidence is, the more accurate the detection result of the corresponding detection object is.
In this embodiment, after the detection result corresponding to the target picture is obtained, whether the detection result meets a preset condition may be further determined, and when the preset condition is met, the target picture is determined to be a negative sample picture.
As an example, the preset condition is that the confidence of any detected object is smaller than a first threshold, where the first threshold may be preset, for example, set to 0.5. Thus, in this example, the determining whether the detection result satisfies the preset condition includes: and judging whether the confidence corresponding to any detection object is smaller than a first threshold value. And when the confidence corresponding to any detection object in the target picture is smaller than a first threshold, indicating that the detection accuracy of the target detection model on the object of the class to which the detection object belongs is low, and determining the target picture as a negative sample picture.
As an example, the preset condition is that the number of the detection objects of the preset type included in the target picture is greater than a second threshold, where the second threshold may be preset according to an empirical value, for example, the second threshold is set to 10; the preset type may be determined according to an application scenario of the target detection model, where the preset type refers to a type of an object that is difficult to recognize in the application scenario of the target detection model, for example, in an unmanned scenario, the preset type may be all types other than a vehicle, such as a bicycle, an electric vehicle, a pedestrian, and the like. It can be understood that, for small objects such as pedestrians, bicycles, motorcycles, etc., when the number of the collected objects is large, the objects are blocked from each other, for example, when a pedestrian crosses a road, the pedestrians are blocked, the bicycles and motorcycles that rush in the morning and evening are also easily blocked, and the target detection model is difficult to detect the blocked objects, which results in missed detection, so that the number of the objects included in the picture and difficult to identify exceeds the second threshold value as a condition for mining the negative sample picture. Thus, in this example, the determining whether the detection result satisfies the preset condition includes: and judging whether the number of the detection objects of the preset type contained in the target picture is greater than a second threshold value. After the detection result corresponding to the target picture is obtained, the number of each type of detection object in the detection result can be counted, and when the number of the preset type of detection objects contained in the target picture is larger than a second threshold value through counting, the target picture is determined to be a negative sample picture.
As an example, the preset condition is that a distance between detection frames corresponding to at least two detection objects of a preset type included in the target picture is smaller than a third threshold, where the distance between the detection frames refers to a distance between two adjacent edges between the detection frames; the third threshold may be preset, and the third threshold may be measured by using pixels, for example, the third threshold may be set to 0, or may be set to a value slightly larger than 0; the preset type can be determined according to the application scene of the target detection model, and the preset type refers to the type of an object difficult to identify in the application scene of the target detection model. It can be understood that when the distances between the detection frames corresponding to the detection objects are smaller, there may be occlusion between the detection objects, for example, when the third threshold is 0, the distances between the detection frames corresponding to at least two detection objects of the preset type are smaller than 0, it may be determined that the detection frames overlap with each other, and the detection objects corresponding to the detection frames are occluded with each other, so that the condition that the distances between the detection frames corresponding to at least two detection objects of the preset type included in the picture are smaller than the third threshold may be taken as the condition for mining the negative sample picture. Thus, in this example, the determining whether the detection result satisfies the preset condition includes: and judging whether the distance between the detection frames corresponding to at least two detection objects of a preset type contained in the target picture is smaller than a third threshold value. And when the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold value, determining that the target picture is a negative sample picture.
It should be noted that each of the preset conditions may be used alone or in combination, for example, the preset condition may be any one of the above conditions, and when the detection result meets the preset condition, the target picture is determined to be a negative sample picture; for another example, the preset condition may include at least two of the above conditions, and when the detection result satisfies any one of the at least two conditions, the target picture is determined to be a negative sample picture, or when the detection result satisfies each of the at least two conditions, the target picture is determined to be a negative sample picture. This is not limited by the present application.
By setting a plurality of conditions, when the detection result is judged to meet any one of the conditions, the detection result is determined to meet the preset condition, the target picture is determined to be the negative sample picture, conditions are provided for realizing automatic excavation of the negative sample picture, the number of the excavated negative sample pictures is increased, and conditions are provided for improving the reliability and accuracy of the target detection model.
And 103, if yes, determining that the target picture is a negative sample picture.
In this embodiment, when the detection result corresponding to the target picture meets the preset condition, the target picture is determined to be a negative sample picture, and a condition is provided for subsequently correcting the target detection model by using the negative sample picture.
In the method for mining the negative sample of the target detection model in this embodiment, the generated target detection model is used to perform target detection on a target picture to be detected, obtain a detection result corresponding to the target picture, judge whether the detection result meets a preset condition, and determine that the target picture is the negative sample picture when the detection result meets the preset condition. Therefore, whether the target picture is the negative sample picture or not is judged by judging whether the detection result meets the preset condition or not, when the detection result meets the preset condition, the target picture is determined to be the negative sample picture, automatic excavation of the negative sample picture is achieved, manual participation is not needed in the process of excavating the negative sample picture, labor cost is saved, compared with a manual excavation mode, the speed of automatically excavating the negative sample picture is high, time consumption is low, excavation efficiency is improved, and conditions are provided for improving reliability and accuracy of a target detection model.
In object detection of a video (e.g., a video stream of 25 frames per second), a plurality of detection objects may be included in the video, and each detection object usually appears continuously, i.e., continuously in a plurality of frames of images of the video stream. In the embodiment of the application, when detecting the detection object in the video, a tracking method may be added, each detection object in the video corresponds to an Identity (ID), and when the ID of one detection object disappears suddenly after appearing several times, it indicates that the detection object loses the tracking, and it may be that the type of the detection object jumps to cause the ID to jump, or that the background object is corrected after being misdetected to cause the ID to disappear. In order to mine a sample containing a false detection object, in a possible implementation manner of the embodiment of the present application, a picture with a small number of occurrences of the detection object may be mined as a negative sample picture, which is described in detail below with reference to fig. 2.
Fig. 2 is a schematic flow chart of a target detection model negative sample mining method according to an embodiment of the second aspect of the present application. As shown in fig. 2, the target detection model negative sample mining method may include the following steps:
Wherein the preset condition comprises at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold.
In this embodiment, for the description of step 201 to step 203, reference may be made to the description of step 101 to step 103 in the foregoing embodiment, and details are not described here again.
And 204, judging whether the detection result contains a target object, wherein the target object is an object detected by the target detection model in N frames of continuous pictures before the target picture is detected, and N is an integer which is greater than 0 and smaller than a fourth threshold value.
The fourth threshold may be preset, for example, the fourth threshold may be set to 5.
In this embodiment, when the detection result does not satisfy the preset condition, it is further determined whether the detection result includes the target object, and when the detection result includes the target object, the target image is considered not to be a negative sample image, and the subsequent operation is continuously performed, that is, step 206 is performed; when the detection result does not include the target object, it indicates that the target object does not appear after N consecutive occurrences, at this time, it may be considered that the category of the target object may be detected incorrectly, or the object in the background is detected incorrectly at the beginning, and then N consecutive pictures including the target object before the target picture may be determined as negative sample pictures, that is, step 205 is performed.
In step 205, N consecutive pictures before the target picture are determined to be negative sample pictures.
In step 206, the subsequent operations are performed.
In this embodiment, when the detection result corresponding to the target picture includes the target object, a subsequent operation is performed, where the subsequent operation may be, for example, performing target detection on a next picture after the target picture, and continuously determining whether the detection result of the next picture meets the condition, and the subsequent operation may also be continuously determining whether the target picture is a negative sample picture by using other methods, for example, continuously determining whether the target object in the target picture appears in a background region of the target picture to determine whether the target picture is a negative sample picture, and so on.
The method for mining negative samples of a target detection model in this embodiment performs target detection on a target picture to be detected by using a generated target detection model, obtains a detection result corresponding to the target picture, determines whether the detection result satisfies a preset condition, determines that the target picture is a negative sample picture when the preset condition is satisfied, further determines whether the detection result includes a target object when the preset condition is not satisfied, wherein the target object is an object detected by the target detection model in N consecutive pictures before the target picture is detected, N is an integer greater than 0 and less than a fourth threshold, and when the target object is not included in the detection result, the N consecutive pictures before the target picture are negative sample pictures, thereby not only mining the negative sample picture when the detection result satisfies the preset condition, but also mining the negative sample picture when the detection result does not satisfy the preset condition, when the detection result does not meet the preset condition, the method further determines whether the detection result contains the target object or not and determines the N frames of continuous pictures before the target picture as the negative sample picture when the detection result does not contain the target object, so that the mining of the false detection sample is realized, and the accuracy of mining the negative sample picture is improved.
When the target detection model is used for carrying out target detection on a target picture, for some detection objects with serious shielding, the shielded detection objects possibly have no detection frame, namely, the shielded detection objects cannot be identified. In addition, for the detection object which is detected by mistake, the detection object which is detected by mistake has higher confidence, so that the sample picture containing the detection object which is detected by mistake cannot be mined out through the confidence of the detection frame. For the problem, in a possible implementation manner of the embodiment of the application, mining of the negative sample picture is supplemented by combining a background modeling technology, so that the negative sample picture including the object which is detected by mistake and detected by omission is mined, and the coverage rate of mining of the negative sample picture is improved. This is described in detail below with reference to fig. 3.
Fig. 3 is a schematic flow chart of a target detection model negative sample mining method according to an embodiment of the third aspect of the present application. As shown in fig. 3, the target detection model negative sample mining method may include the following steps:
Wherein the preset condition comprises at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold.
In this embodiment, for the description of steps 301 to 303, reference may be made to the description of steps 101 to 103 in the foregoing embodiment, and details are not described here again.
The background recognition model can be obtained by pre-training, and the background recognition model can recognize a foreground region and a background region of the picture. The foreground region is a region where a moving object is located, and the background recognition model recognizes the foreground region and the background region, namely, the moving object and the static object in the picture. The foreground region can be a whole, or a region where each moving object is located in the picture can be identified as a foreground region, and when a plurality of moving objects exist in the picture, the background identification model identifies a plurality of foreground regions.
In this embodiment, when the detection result does not satisfy the preset condition, the preset background recognition model may be further used to recognize the target picture, so as to determine a foreground region and a background region included in the target picture.
In this embodiment, after the background region and the foreground region in the target picture are identified, whether the background region contains the detection object may be determined according to the detection result of the target picture.
It can be understood that when a certain detection object appears in the background region of the picture, it indicates that there are two possibilities for the object, one is a moving object that is stationary for a long time, and the other is a stationary object that is erroneously detected as a moving object, and both of these two cases indicate that the detection result corresponding to the target picture is erroneously detected, so in this embodiment, whether the detection object is included in the background region can be used as a condition for determining whether the target picture is a negative sample picture.
In this embodiment, when it is known that the background region contains the detection object, the target picture is determined to be a negative sample picture, that is, step 307 is executed; when it is known that the bit in the background region contains the detection object, the subsequent operation is performed, i.e., step 308 is performed.
And step 306, judging whether the foreground area contains the detection object or not according to the detection result corresponding to the target picture.
In this embodiment, after the background region and the foreground region in the target picture are identified, whether the foreground region includes the detection object may be determined according to the detection result of the target picture.
The background recognition model can recognize a foreground region and a background region, namely, recognize all moving objects in the target picture, judge whether the foreground region contains a detection object or not by comparing whether the foreground region contains the moving object without a detection frame or not according to a detection result corresponding to the target picture, and determine that the foreground region does not contain the detection object if the foreground region contains the moving object without the detection frame, thereby determining that the target picture is a negative sample picture, namely, executing step 307; if all the moving objects in the foreground region contain the detection frame, the following operation is performed, i.e., step 308 is performed.
For example, when the foreground region is a whole and there is a moving object without a detection frame in the foreground region, it is determined that the foreground region does not include a detection object, and the corresponding target picture is determined as a negative sample picture; and when the foreground region is a small region corresponding to a moving object and does not contain a detection frame, determining that the foreground region does not contain a detection object, and determining the corresponding target picture as a negative sample picture.
In step 308, the subsequent operations are performed.
In this embodiment, when it is known that the background region of the target picture does not include the detection object, or when it is known that the foreground region of the target picture includes the detection object, a subsequent operation is performed, where the subsequent operation may be, for example, performing target detection on a next picture after the target picture, and continuously determining whether a detection result of the next picture meets a condition, or other subsequent operations, which is not limited in this application.
The method for mining negative samples of a target detection model in this embodiment performs target detection on a target picture to be detected by using a generated target detection model, obtains a detection result corresponding to the target picture, determines whether the detection result satisfies a preset condition, determines that the target picture is a negative sample picture when the preset condition is satisfied, further identifies the target picture by using a preset background identification model when the preset condition is not satisfied, to determine a foreground region and a background region included in the target picture, determines whether a detection object is included in the background region according to the detection result corresponding to the target picture, determines that the target picture is a negative sample picture if the detection object is included, or determines whether the detection object is included in the foreground region according to the detection result corresponding to the target picture, and determines that the target picture is a negative sample picture if the detection object is not included, thereby not only realizing mining of the negative sample picture when the detection result satisfies the preset condition, the mining of the negative sample picture when the detection result does not meet the preset condition is also realized, the negative sample picture is further mined by combining the identification result of the background identification model when the detection result does not meet the preset condition, and the target picture which does not contain the detection object in the foreground area is determined as the negative sample picture, so that the mining of the sample picture containing the detection object with weaker detection capability is realized, and the coverage rate and the accuracy of the mining of the negative sample picture are improved.
Further, in a possible implementation manner of the embodiment of the application, after the negative sample picture is mined, the generated target detection model may be modified and trained by using the determined negative sample picture, so as to improve the recognition accuracy of the target detection model on the negative sample picture, and thus improve the reliability and accuracy of the target detection model.
According to the embodiment of the application, the application also provides a target detection model negative sample mining device.
Fig. 4 is a schematic structural diagram of a target detection model negative sample mining device according to a fourth embodiment of the present application. As shown in fig. 4, the target detection model negative sample mining 40 includes: an obtaining module 410, a first judging module 420, and a determining module 430.
The obtaining module 410 is configured to perform target detection on a target picture to be detected by using the generated target detection model, and obtain a detection result corresponding to the target picture.
The first determining module 420 is configured to determine whether the detection result meets a preset condition, where the preset condition includes at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold.
The determining module 430 is configured to determine that the target picture is a negative sample picture when the detection result meets a preset condition.
Further, in a possible implementation manner of the embodiment of the present application, as shown in fig. 5, on the basis of the embodiment shown in fig. 4, the target detection model negative sample mining 40 further includes:
the second determining module 440 is configured to determine whether the detection result includes a target object when the detection result does not satisfy a preset condition, where the target object is an object detected by the target detection model in N consecutive pictures before the target picture is detected, and N is an integer greater than 0 and smaller than a fourth threshold.
In this embodiment, the determining module 430 is further configured to: and if the detection result does not contain the target object, determining that the N frames of continuous pictures before the target picture are negative sample pictures.
When the detection result does not meet the preset condition, the method further determines whether the detection result contains the target object or not and determines the N frames of continuous pictures before the target picture as the negative sample picture when the detection result does not contain the target object, so that the mining of the false detection sample is realized, and the accuracy of mining the negative sample picture is improved.
In a possible implementation manner of the embodiment of the present application, as shown in fig. 6, on the basis of the embodiment shown in fig. 4, the target detection model negative sample mining 40 further includes:
the processing module 450 is configured to identify the target picture by using a preset background identification model when the detection result does not meet a preset condition, so as to determine a foreground region and a background region included in the target picture.
The third determining module 460 is configured to determine whether the background region includes the detected object according to the detection result corresponding to the target picture.
In this embodiment, the determining module 430 is further configured to: and if the background area contains the detection object, determining that the target picture is a negative sample picture.
When the detection result does not meet the preset condition, the negative sample picture is further mined by combining the recognition result of the background recognition model, and the target picture containing the detection object in the background area is determined as the negative sample picture, so that the mining of the false detection sample is realized, and the coverage rate and the accuracy of the mining of the negative sample picture are improved.
In a possible implementation manner of the embodiment of the present application, the third determining module 460 is further configured to: and judging whether the foreground region contains the detection object or not according to the detection result corresponding to the target picture. The determining module 430 is further configured to: and if the foreground region does not contain the detection object, determining that the target picture is a negative sample picture.
When the detection result does not meet the preset condition, the negative sample picture is further mined by combining the identification result of the background identification model, and the target picture which does not contain the detection object in the foreground area is determined as the negative sample picture, so that the mining of the sample picture containing the detection object with weaker detection capability is realized, and the coverage rate and the accuracy of the mining of the negative sample picture are improved.
Further, in a possible implementation manner of the embodiment of the present application, on the basis of the foregoing embodiment, the target detection model negative sample mining 40 further includes: and the correction module is used for performing correction training on the generated target detection model by using the determined negative sample picture.
For example, as shown in fig. 7, based on the embodiment shown in fig. 4, the target detection model negative sample mining 40 further includes: and a correction module 470.
The target detection model is corrected and trained by using the determined negative sample picture, so that the reliability and the accuracy of the target detection model are improved.
It should be noted that the foregoing explanation of the embodiment of the target detection model negative sample mining method is also applicable to the target detection model negative sample mining apparatus in the embodiment of the present application, and the implementation principle is similar, and is not described herein again.
According to the target detection model negative sample mining device, the generated target detection model is utilized, the target picture to be detected is subjected to target detection, the detection result corresponding to the target picture is obtained, whether the detection result meets the preset condition or not is judged, and when the detection result meets the preset condition, the target picture is determined to be the negative sample picture. Therefore, whether the target picture is the negative sample picture or not is judged by judging whether the detection result meets the preset condition or not, when the detection result meets the preset condition, the target picture is determined to be the negative sample picture, automatic excavation of the negative sample picture is achieved, manual participation is not needed in the process of excavating the negative sample picture, labor cost is saved, compared with a manual excavation mode, the speed of automatically excavating the negative sample picture is high, time consumption is low, excavation efficiency is improved, and conditions are provided for improving reliability and accuracy of a target detection model.
According to an embodiment of the present application, an electronic device and a readable storage medium are also provided.
Fig. 8 is a block diagram of an electronic device of a target detection model negative sample mining method according to an embodiment of the present application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application that are described and/or claimed herein.
As shown in fig. 8, the electronic apparatus includes: one or more processors 701, a memory 702, and interfaces for connecting the various components, including a high-speed interface and a low-speed interface. The various components are interconnected using different buses and may be mounted on a common motherboard or in other manners as desired. The processor 701 may process instructions for execution within the electronic device, including instructions stored in or on the memory 702 to display graphical information of a GUI on an external input/output apparatus (such as a display device coupled to an interface). In other embodiments, multiple processors and/or multiple buses may be used, along with multiple memories and multiple memories, as desired. Also, multiple electronic devices may be connected, with each device providing portions of the necessary operations (e.g., as a server array, a group of blade servers, or a multi-processor system). In fig. 8, one processor 701 is taken as an example.
The memory 702 is a non-transitory computer readable storage medium as provided herein. The memory 702 stores instructions executable by at least one processor to cause the at least one processor 701 to perform the target detection model negative sample mining method provided herein. A non-transitory computer-readable storage medium of the present application stores computer instructions for causing a computer to perform the target detection model negative sample mining method provided herein.
The memory 702, which is a non-transitory computer-readable storage medium, may be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as program instructions/modules corresponding to the target detection model negative sample mining method in the embodiments of the present application (e.g., the obtaining module 410, the first determining module 420, and the determining module 440 shown in fig. 6). The processor 701 executes various functional applications of the server and data processing by running non-transitory software programs, instructions, and modules stored in the memory 702, that is, implements the target detection model negative sample mining method in the above-described method embodiments.
The memory 702 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created from use of an electronic device that performs the target detection model negative sample mining method, and the like. Further, the memory 702 may include high speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 702 may optionally include memory located remotely from the processor 701, and such remote memory may be connected over a network to an electronic device that performs the target detection model negative sample mining method. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The electronic device executing the target detection model negative sample mining method may further include: an input device 703 and an output device 704. The processor 701, the memory 702, the input device 703 and the output device 704 may be connected by a bus or other means, and fig. 8 illustrates an example of a connection by a bus.
The input device 703 may receive input numeric or character information and generate key signal inputs related to user settings and function control of an electronic apparatus performing the target detection model negative sample mining method, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, a pointing stick, one or more mouse buttons, a track ball, a joystick, or other input devices. The output devices 704 may include a display device, auxiliary lighting devices (e.g., LEDs), and tactile feedback devices (e.g., vibrating motors), among others. The display device may include, but is not limited to, a Liquid Crystal Display (LCD), a Light Emitting Diode (LED) display, and a plasma display. In some implementations, the display device can be a touch screen.
Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, integrated circuitry, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
These computer programs (also known as programs, software applications, or code) include machine instructions for a programmable processor, and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.
The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present application may be executed in parallel, sequentially, or in different orders, and the present invention is not limited thereto as long as the desired results of the technical solutions disclosed in the present application can be achieved.
The above-described embodiments should not be construed as limiting the scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (12)
1. A target detection model negative sample mining method is characterized by comprising the following steps:
carrying out target detection on a target picture to be detected by using the generated target detection model to obtain a detection result corresponding to the target picture;
judging whether the detection result meets a preset condition or not, wherein the preset condition comprises at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold;
and if so, determining that the target picture is a negative sample picture.
2. The method of claim 1, wherein after determining whether the detection result satisfies a predetermined condition, the method further comprises:
if not, judging whether the detection result contains a target object, wherein the target object is an object detected by the target detection model in N frames of continuous pictures before the target picture is detected, and N is an integer which is greater than 0 and smaller than a fourth threshold value;
and if not, determining that the N frames of continuous pictures before the target picture are negative sample pictures.
3. The method of claim 1, wherein after determining whether the detection result satisfies a predetermined condition, the method further comprises:
if the current image is not met, identifying the target image by using a preset background identification model so as to determine a foreground area and a background area contained in the target image;
judging whether the background region contains a detection object or not according to a detection result corresponding to the target picture;
and if so, determining that the target picture is a negative sample picture.
4. The method of claim 3, wherein after determining the foreground region and the background region included in the target picture, further comprising:
judging whether the foreground region contains a detection object or not according to a detection result corresponding to the target picture;
if not, determining that the target picture is a negative sample picture.
5. The method of any of claims 1-4, wherein after determining that the target picture is a negative sample picture, further comprising:
and performing correction training on the generated target detection model by using the determined negative sample picture.
6. A target detection model negative sample excavating device is characterized by comprising:
the acquisition module is used for carrying out target detection on a target picture to be detected by utilizing the generated target detection model and acquiring a detection result corresponding to the target picture;
the first judging module is configured to judge whether the detection result meets a preset condition, where the preset condition includes at least one of the following conditions: the confidence of any detection object is smaller than a first threshold, the number of the detection objects of the preset type contained in the target picture is larger than a second threshold, and the distance between the detection frames corresponding to at least two detection objects of the preset type contained in the target picture is smaller than a third threshold;
and the determining module is used for determining that the target picture is a negative sample picture when the detection result meets a preset condition.
7. The apparatus of claim 6, further comprising:
a second determining module, configured to determine whether a target object is included in the detection result when the detection result does not meet a preset condition, where the target object is an object detected by the target detection model in N consecutive pictures before the target picture is detected, and N is an integer greater than 0 and smaller than a fourth threshold;
the determining module is further configured to:
and if the detection result does not contain the target object, determining that the N frames of continuous pictures before the target picture are negative sample pictures.
8. The apparatus of claim 6, further comprising:
the processing module is used for identifying the target picture by using a preset background identification model when the detection result does not meet a preset condition so as to determine a foreground area and a background area contained in the target picture;
the third judging module is used for judging whether the background area contains a detection object or not according to the detection result corresponding to the target picture;
the determining module is further configured to:
and if the background region contains the detection object, determining that the target picture is a negative sample picture.
9. The apparatus of claim 8, wherein the third determining module is further configured to:
judging whether the foreground region contains a detection object or not according to a detection result corresponding to the target picture;
the determining module is further configured to:
and if the foreground region does not contain the detection object, determining that the target picture is a negative sample picture.
10. The apparatus of any of claims 6-9, further comprising:
and the correction module is used for performing correction training on the generated target detection model by using the determined negative sample picture.
11. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the target detection model negative sample mining method of any of claims 1-5.
12. A non-transitory computer readable storage medium having stored thereon computer instructions for causing a computer to execute the target detection model negative sample mining method of any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911136309.2A CN110968718B (en) | 2019-11-19 | 2019-11-19 | Target detection model negative sample mining method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911136309.2A CN110968718B (en) | 2019-11-19 | 2019-11-19 | Target detection model negative sample mining method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110968718A true CN110968718A (en) | 2020-04-07 |
| CN110968718B CN110968718B (en) | 2023-07-14 |
Family
ID=70030929
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911136309.2A Active CN110968718B (en) | 2019-11-19 | 2019-11-19 | Target detection model negative sample mining method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110968718B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784095A (en) * | 2021-01-18 | 2021-05-11 | 北京洛塔信息技术有限公司 | Difficult sample data mining method, device, equipment and storage medium |
| CN113344086A (en) * | 2021-06-16 | 2021-09-03 | 深圳市商汤科技有限公司 | Man-machine loop method, device, system, electronic equipment and storage medium |
| CN113454649A (en) * | 2021-06-17 | 2021-09-28 | 商汤国际私人有限公司 | Target detection method, target detection device, electronic equipment and computer-readable storage medium |
| CN113627229A (en) * | 2021-05-31 | 2021-11-09 | 中国兵器工业计算机应用技术研究所 | Object detection method, system, device and computer storage medium |
| WO2022027913A1 (en) * | 2020-08-05 | 2022-02-10 | 深圳市优必选科技股份有限公司 | Target detection model generating method and apparatus, device and storage medium |
| CN116450808A (en) * | 2023-06-15 | 2023-07-18 | 腾讯科技（深圳）有限公司 | Data processing method and device and storage medium |
| WO2023184833A1 (en) * | 2022-03-28 | 2023-10-05 | 上海商汤智能科技有限公司 | Detection result processing method and apparatus, device, medium, and computer program product |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103366163A (en) * | 2013-07-15 | 2013-10-23 | 北京丰华联合科技有限公司 | Human face detection system and method based on incremental learning |
| CN104573669A (en) * | 2015-01-27 | 2015-04-29 | 中国科学院自动化研究所 | Image object detection method |
| US20160155241A1 (en) * | 2013-06-17 | 2016-06-02 | Huawei Device Co., Ltd. | Target Detection Method and Apparatus Based On Online Training |
| CN107886495A (en) * | 2017-09-30 | 2018-04-06 | 北京得华机器人技术研究院有限公司 | A kind of auto-parts defect identification method based on similarity mode |
| CN108122247A (en) * | 2017-12-25 | 2018-06-05 | 北京航空航天大学 | A kind of video object detection method based on saliency and feature prior model |
| US20180174046A1 (en) * | 2016-12-15 | 2018-06-21 | Beijing Kuangshi Technology Co., Ltd. | Target detection method and device, neural network training method and device |
| CN108416287A (en) * | 2018-03-04 | 2018-08-17 | 南京理工大学 | A kind of pedestrian detection method excavated based on omission negative sample |
| US20190138791A1 (en) * | 2016-08-10 | 2019-05-09 | Tencent Technology (Shenzhen) Company Limited | Key point positioning method, terminal, and computer storage medium |
| CN109931946A (en) * | 2019-04-10 | 2019-06-25 | 福州大学 | Blind visual range-finding navigation method based on Android intelligent |
| CN109937152A (en) * | 2017-08-10 | 2019-06-25 | 北京市商汤科技开发有限公司 | Driving condition supervision method and apparatus, driver's monitoring system, vehicle |
| CN110443280A (en) * | 2019-07-05 | 2019-11-12 | 北京达佳互联信息技术有限公司 | Training method, device and the storage medium of image detection model |
-
2019
- 2019-11-19 CN CN201911136309.2A patent/CN110968718B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160155241A1 (en) * | 2013-06-17 | 2016-06-02 | Huawei Device Co., Ltd. | Target Detection Method and Apparatus Based On Online Training |
| CN103366163A (en) * | 2013-07-15 | 2013-10-23 | 北京丰华联合科技有限公司 | Human face detection system and method based on incremental learning |
| CN104573669A (en) * | 2015-01-27 | 2015-04-29 | 中国科学院自动化研究所 | Image object detection method |
| US20190138791A1 (en) * | 2016-08-10 | 2019-05-09 | Tencent Technology (Shenzhen) Company Limited | Key point positioning method, terminal, and computer storage medium |
| US20180174046A1 (en) * | 2016-12-15 | 2018-06-21 | Beijing Kuangshi Technology Co., Ltd. | Target detection method and device, neural network training method and device |
| CN109937152A (en) * | 2017-08-10 | 2019-06-25 | 北京市商汤科技开发有限公司 | Driving condition supervision method and apparatus, driver's monitoring system, vehicle |
| CN107886495A (en) * | 2017-09-30 | 2018-04-06 | 北京得华机器人技术研究院有限公司 | A kind of auto-parts defect identification method based on similarity mode |
| CN108122247A (en) * | 2017-12-25 | 2018-06-05 | 北京航空航天大学 | A kind of video object detection method based on saliency and feature prior model |
| CN108416287A (en) * | 2018-03-04 | 2018-08-17 | 南京理工大学 | A kind of pedestrian detection method excavated based on omission negative sample |
| CN109931946A (en) * | 2019-04-10 | 2019-06-25 | 福州大学 | Blind visual range-finding navigation method based on Android intelligent |
| CN110443280A (en) * | 2019-07-05 | 2019-11-12 | 北京达佳互联信息技术有限公司 | Training method, device and the storage medium of image detection model |
Non-Patent Citations (2)
| Title |
|---|
| WEI DONG ET AL.: "High-Confidence Near-Duplicate Image Detection", ACM * |
| 刘芷含等: "遗漏负样本挖掘的行人检测方法", 计算机与数字工程 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022027913A1 (en) * | 2020-08-05 | 2022-02-10 | 深圳市优必选科技股份有限公司 | Target detection model generating method and apparatus, device and storage medium |
| CN112784095A (en) * | 2021-01-18 | 2021-05-11 | 北京洛塔信息技术有限公司 | Difficult sample data mining method, device, equipment and storage medium |
| CN113627229A (en) * | 2021-05-31 | 2021-11-09 | 中国兵器工业计算机应用技术研究所 | Object detection method, system, device and computer storage medium |
| CN113627229B (en) * | 2021-05-31 | 2023-08-25 | 中国兵器工业计算机应用技术研究所 | Target detection method, system, device and computer storage medium |
| CN113344086A (en) * | 2021-06-16 | 2021-09-03 | 深圳市商汤科技有限公司 | Man-machine loop method, device, system, electronic equipment and storage medium |
| CN113344086B (en) * | 2021-06-16 | 2022-07-01 | 深圳市商汤科技有限公司 | Man-machine loop method, device, system, electronic equipment and storage medium |
| CN113454649A (en) * | 2021-06-17 | 2021-09-28 | 商汤国际私人有限公司 | Target detection method, target detection device, electronic equipment and computer-readable storage medium |
| CN113454649B (en) * | 2021-06-17 | 2024-05-24 | 商汤国际私人有限公司 | Target detection method, apparatus, electronic device, and computer-readable storage medium |
| WO2023184833A1 (en) * | 2022-03-28 | 2023-10-05 | 上海商汤智能科技有限公司 | Detection result processing method and apparatus, device, medium, and computer program product |
| CN116450808A (en) * | 2023-06-15 | 2023-07-18 | 腾讯科技（深圳）有限公司 | Data processing method and device and storage medium |
| CN116450808B (en) * | 2023-06-15 | 2023-09-19 | 腾讯科技（深圳）有限公司 | Data processing method and device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110968718B (en) | 2023-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110968718A (en) | Target detection model negative sample mining method and device and electronic equipment | |
| CN111723768B (en) | Method, device, equipment and storage medium for vehicle re-identification | |
| CN112528786B (en) | Vehicle tracking method and device and electronic equipment | |
| CN111709328A (en) | Vehicle tracking method and device and electronic equipment | |
| CN111275011B (en) | Mobile traffic light detection method and device, electronic equipment and storage medium | |
| CN110796865B (en) | Intelligent traffic control method and device, electronic equipment and storage medium | |
| CN111292531A (en) | Tracking method, device and equipment of traffic signal lamp and storage medium | |
| CN110717933B (en) | Post-processing method, device, equipment and medium for moving object missed detection | |
| JP7200207B2 (en) | Map generation method, map generation device, electronic device, non-transitory computer-readable storage medium and computer program | |
| CN110659600A (en) | Object detection method, device and equipment | |
| CN111291681A (en) | Method, device and equipment for detecting lane line change information | |
| CN111652112A (en) | Lane flow direction identification method and device, electronic equipment and storage medium | |
| CN112257604A (en) | Image detection method, image detection device, electronic equipment and storage medium | |
| CN111444819B (en) | Cut frame determining method, network training method, device, equipment and storage medium | |
| CN111191619A (en) | Method, device and equipment for detecting virtual line segment of lane line and readable storage medium | |
| CN111601013A (en) | Method and apparatus for processing video frames | |
| CN113657398B (en) | Image recognition method and device | |
| CN112560772B (en) | Face recognition method, device, equipment and storage medium | |
| CN113361303B (en) | Temporary traffic sign board identification method, device and equipment | |
| CN111401248B (en) | Sky area identification method and device, electronic equipment and storage medium | |
| CN110796864B (en) | Intelligent traffic control method, device, electronic equipment and storage medium | |
| CN113011298A (en) | Truncated object sample generation method, target detection method, road side equipment and cloud control platform | |
| CN111027195B (en) | Simulation scene generation method, device and equipment | |
| CN111460072B (en) | Lane line detection method, device, equipment and storage medium | |
| CN110798681A (en) | Monitoring method and device of imaging equipment and computer equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |