CN111310660A - Target detection false alarm suppression method and device for ADAS scene - Google Patents
Target detection false alarm suppression method and device for ADAS scene Download PDFInfo
- Publication number
- CN111310660A CN111310660A CN202010095000.XA CN202010095000A CN111310660A CN 111310660 A CN111310660 A CN 111310660A CN 202010095000 A CN202010095000 A CN 202010095000A CN 111310660 A CN111310660 A CN 111310660A
- Authority
- CN
- China
- Prior art keywords
- data set
- sample data
- target detection
- negative sample
- false alarm
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 121
- 238000000034 method Methods 0.000 title claims abstract description 58
- 230000001629 suppression Effects 0.000 title claims abstract description 37
- 102100034112 Alkyldihydroxyacetonephosphate synthase, peroxisomal Human genes 0.000 title claims abstract 11
- 101000799143 Homo sapiens Alkyldihydroxyacetonephosphate synthase, peroxisomal Proteins 0.000 title claims abstract 11
- 238000000848 angular dependent Auger electron spectroscopy Methods 0.000 title claims abstract 11
- 238000012549 training Methods 0.000 claims abstract description 93
- 238000005070 sampling Methods 0.000 claims abstract description 31
- 238000003860 storage Methods 0.000 claims abstract description 17
- 238000005457 optimization Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 5
- 230000000295 complement effect Effects 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 22
- 238000005516 engineering process Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 abstract description 3
- 230000002401 inhibitory effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 28
- 238000010586 diagram Methods 0.000 description 11
- 238000012545 processing Methods 0.000 description 8
- 238000004590 computer program Methods 0.000 description 7
- 238000012795 verification Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000005764 inhibitory process Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a target detection false alarm suppression method and device for an ADAS scene, electronic equipment and a readable storage medium. The method comprises the following steps: collecting a positive sample data set and a negative sample data set in an ADAS scene to serve as a first training set; sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain a second training set; inputting the first training set and the second training set into a preset network model for training to obtain a target detection model; and optimizing the target detection model based on a preset loss function so as to carry out false alarm suppression of target detection according to the optimized target detection model. The method and the device solve the technical problem that a target detection algorithm for the ADAS scene in the related technology cannot effectively inhibit false alarm, so that the detection precision is not high. By the method and the device, the aim of effectively inhibiting the target detection false alarm is fulfilled, and the technical effect of improving the detection precision of the target detection algorithm is achieved.
Description
Technical Field
The application relates to the field of auxiliary driving, in particular to a target detection false alarm suppression method and device for an ADAS scene, electronic equipment and a readable storage medium.
Background
An Advanced Driving Assistance System (ADAS) is a System that uses various sensors (millimeter wave radar, laser radar, single/binocular camera and satellite navigation) installed on a vehicle to sense the surrounding environment at any time during the Driving process of the vehicle, collect data, identify, detect and track static and dynamic objects, and combine with the map data of a navigator to perform systematic calculation and analysis, thereby enabling drivers to detect the danger that may occur in advance and effectively increasing the comfort and safety of the Driving of the vehicle.
In Advanced Driver Assistance Systems (ADAS) scenarios, detection of objects such as vehicles, pedestrians, traffic signs, etc. is one of the core tasks. With the application development of artificial intelligence in the field of automobiles, the advancement and reliability of the ADAS target detection technology make a breakthrough progress, and the detection accuracy of the ADAS system can be greatly improved by constructing a Deep Convolutional Neural Network (DCNN) model and adjusting target detection algorithm parameters.
In the current practical application scenario, a target detection algorithm based on the DCNN gives a boundary box to describe the position of a target according to the ADAS scene requirements, and classifies the target in the boundary box. The method can be divided into a two-stage algorithm and a one-stage algorithm according to a detection process: in a two-stage target detection algorithm represented by fast-RCNN, a target candidate frame is generated by a Region Proposed Network (RPN) method and the like, and then the selected candidate frame is subjected to bounding box regression and classification. In a one-stage target detection algorithm represented by Yolo and SSD, the bounding box information and the classification information of a target are fused into a DCNN model together, and border regression and classification are directly performed. In a target detection algorithm based on DCNN, a detected target is generally divided into a Positive sample (Positive) and a Negative sample (Negative) according to an anchor point, and after a series of algorithm operations, the more Positive samples (TP) that are finally output as correct detection, the stronger the detection capability of target detection, and the higher the accuracy. Meanwhile, the more targets (FP) that are wrongly classified as Positive samples, the higher the False detection rate of target detection, and the more False alarms (False information that a signal comes when no signal comes) are generated in the ADAS scene.
However, the inventor finds that the ADAS system in the related art has high requirements on detection real-time performance, robustness and false detection rate, and it is difficult to construct a target detection backbone network structure of a complex network structure while meeting the real-time performance, so that by means of adjusting anchor parameters, constructing a feature pyramid network structure and the like, a high recognition rate can be achieved in an ADAS scene, and the robustness requirement is met, but at the same time, a high false detection rate is easily caused, and more false alarms are generated, which interferes with normal driving of a driver. The common false alarm suppression method can establish a positive and negative sample library by collecting pictures containing detection targets and pictures not containing detection targets in a large number of real scenes, but the ADAS actual application scene is often complex, the construction of negative samples is easy to generate deviation, the method possibly causes extremely unbalanced division of the positive and negative samples in an algorithm layer, the DCNN training process is difficult to converge, the detection accuracy is easy to lose, and especially the detection of small and medium targets can be influenced.
Aiming at the problem that the target detection algorithm for the ADAS scene in the related technology cannot effectively inhibit false alarm to cause low detection precision, an effective solution is not provided at present.
Disclosure of Invention
The present application mainly aims to provide a method and an apparatus for suppressing a target detection false alarm for an ADAS scene, an electronic device, and a readable storage medium, so as to solve the problem in the related art that a target detection algorithm for an ADAS scene is not high in detection accuracy due to the fact that a false alarm cannot be effectively suppressed.
To achieve the above object, according to a first aspect of the present application, a target detection false alarm suppression method for an ADAS scenario is provided.
The target detection false alarm suppression method for the ADAS scene comprises the following steps: collecting a positive sample data set and a negative sample data set in an ADAS scene to serve as a first training set; sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain a second training set; inputting the first training set and the second training set into a preset network model for training to obtain a target detection model; and optimizing the target detection model based on a preset loss function so as to carry out false alarm suppression of target detection according to the optimized target detection model.
Further, the ADAS scenario includes at least one sub-application scenario, and the acquiring a positive sample data set and a negative sample data set in the ADAS scenario to serve as a first training set includes: acquiring a positive sample data set, a negative sample data set and a false alarm image data set in each sub-application scene; classifying the positive sample data set, the negative sample data set and the false alarm image data set in each sub-application scene according to preset classification items to obtain a training set in each sub-application scene, wherein the preset classification items include but are not limited to time, weather, climate, road environment and image quality.
Further, the sampling and randomly padding the negative sample data set into the positive sample data set to obtain a second training set includes: constructing a negative sample region clipping queue, and updating the negative sample region clipping queue in real time; sampling the negative sample region clipping queue, and randomly filling the sampled negative sample region into a positive sample region in the positive sample data set; and randomly removing the positive sample marking frame corresponding to the positive sample region to obtain the second training set.
Further, the preset loss function is composed of the following equation:
wherein the content of the first and second substances,
i is the current iterationNumber of times, ImaxAs a result of the total number of iterations,
in order to be an illustrative function of the system,
is LdetThe complement of the class prediction is selected,
wherein the content of the first and second substances,
n is the number of anchor point predictions as positive samples, LconfAs a function of classification loss, LlocFor bounding box loss function, α is the reconciliation parameter between the classification and the bounding box.
Further, the optimizing the target detection model based on the preset loss function to perform false alarm suppression of target detection according to the optimized target detection model includes: sampling the first training set and the second training set to obtain a sampled data set; mixing the sampling data set and the negative sample data set according to a preset proportion to obtain a mixed data set; and fine-tuning the target detection model according to the mixed data set.
In order to achieve the above object, according to a second aspect of the present application, there is provided an object detection false alarm suppression apparatus for an ADAS scene.
The target detection false alarm suppression device for the ADAS scene comprises: the acquisition module is used for acquiring a positive sample data set and a negative sample data set in an ADAS scene to serve as a first training set; the increment module is used for sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain a second training set; the training module is used for inputting the first training set and the second training set into a preset network model for training so as to obtain a target detection model; and the optimization module is used for optimizing the target detection model based on a preset loss function so as to carry out false alarm suppression of target detection according to the optimized target detection model.
Further, the ADAS scene includes at least one sub-application scene, and the acquisition module includes: the acquisition unit is used for acquiring a positive sample data set, a negative sample data set and a false alarm image data set in each sub-application scene; and the classification unit is used for classifying the positive sample data set, the negative sample data set and the false alarm image data set in each sub-application scene according to preset classification items to obtain a training set in each sub-application scene, wherein the preset classification items include but are not limited to time, weather, climate, road environment and image quality.
Further, the increment module includes: the construction unit is used for constructing a negative sample region clipping queue and updating the negative sample region clipping queue in real time; the sampling unit is used for sampling the negative sample region clipping queue and randomly filling the sampled negative sample region into the positive sample region in the positive sample data set; and the removing unit is used for randomly removing the positive sample marking frame corresponding to the positive sample region to obtain the second training set.
In order to achieve the above object, according to a third aspect of the present application, there is provided an electronic apparatus comprising: one or more processors; storage means for storing one or more programs; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement a method as in any preceding item
To achieve the above object, according to a fourth aspect of the present application, there is provided a non-transitory readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method according to any of the preceding claims.
In the embodiment of the application, a positive sample data set and a negative sample data set under an ADAS scene are collected to be used as a first training set; sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain a second training set; the method comprises the steps of inputting a first training set and a second training set into a preset network model for training to obtain a target detection model, optimizing the target detection model based on a preset loss function to carry out false alarm suppression of target detection according to the optimized target detection model, and achieving the purpose of effectively suppressing the target detection false alarm, so that the technical effect of improving the detection precision of a target detection algorithm is achieved, and the technical problem that the detection precision is not high due to the fact that the false alarm cannot be effectively suppressed in the target detection algorithm for an ADAS scene in the related technology is solved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, serve to provide a further understanding of the application and to enable other features, objects, and advantages of the application to be more apparent. The drawings and their description illustrate the embodiments of the invention and do not limit it. In the drawings:
fig. 1 is a schematic flow diagram of a target detection false alarm suppression method for an ADAS scenario according to a first embodiment of the present application;
fig. 2 is a schematic flow chart of a target detection false alarm suppression method for an ADAS scenario according to a second embodiment of the present application;
fig. 3 is a schematic flow chart of a target detection false alarm suppression method for ADAS scenarios according to a third embodiment of the present application;
fig. 4 is a schematic flow chart of a target detection false alarm suppression method for an ADAS scene according to a fourth embodiment of the present application;
fig. 5 is a schematic structural diagram of a target detection false alarm suppression apparatus for an ADAS scene according to an embodiment of the present application; and
fig. 6 is a schematic diagram of a composition structure of an electronic device according to an embodiment of the present application.
Detailed Description
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only partial embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It should be understood that the data so used may be interchanged under appropriate circumstances such that embodiments of the application described herein may be used. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
According to an embodiment of the present invention, there is provided a target detection false alarm suppression method for an ADAS scene, as shown in fig. 1, the method includes the following steps S101 to S104:
step S101, collecting a positive sample data set and a negative sample data set in an ADAS scene to serve as a first training set.
The ADAS scene is widely applied, and the construction of a data set covering the whole is the premise of realizing target detection optimization and effectively inhibiting false alarms. Therefore, in the specific implementation of the embodiment of the application, the positive sample data set and the negative sample data set in the ADAS scene are collected at first, and the collected positive sample data set and the collected negative sample data set in the ADAS scene are divided into the training set and the verification set according to a certain proportion. Specifically, the images in the data set may or may not include positive sample information of vehicles, pedestrians, and the like, and it is necessary to try to keep the distribution of the images consistent with the actual application scene. And finally, independently adding image data containing false alarms in the training set and the verification set, wherein the partial data can be added as a class of the data set, and the single class of positive sample data is kept at the same order of magnitude.
And S102, sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain a second training set.
In a target detection algorithm based on a Deep Convolutional Neural Network (DCNN), positive and negative samples are mainly distinguished through anchor points of an algorithm layer, a series of data increment means are adopted in an actual training process to strive for a data set to cover more actual scenes, the data increment means usually only aim at real positive samples, the increment means usually performs brightness, saturation, contrast, clipping, deformation and other methods on an image containing the positive samples, and the increment means is not beneficial to the inhibition of false alarms. Therefore, for the false alarm situation in the ADAS scenario, the embodiment of the present application provides a data increment strategy for false alarm suppression, which first needs to randomly sample the acquired negative sample data set, and randomly fill the sampled negative sample into the positive sample to serve as a new training set.
Step S103, inputting the first training set and the second training set into a preset network model for training to obtain a target detection model.
In specific implementation, the data in the second training set obtained after random filling and the data in the normally collected first training set are mixed and input into a preset network model for training, and then a preliminary target detection model is obtained.
And step S104, optimizing the target detection model based on a preset loss function so as to carry out false alarm suppression of target detection according to the optimized target detection model.
In the target detection algorithm, each anchor point respectively generates classification information and frame information of a target, positive and negative samples are distinguished through classification and frames, and the purpose of training is to draw the classification information and the frame information predicted by the anchor points to the real classification information and the frame information. The method and the device for detecting the false alarm further optimize the target detection model obtained after training by adopting the preset loss function oriented to the false alarm suppression, and the optimized target detection model can be beneficial to the suppression of the false alarm.
As a preferred implementation manner of the embodiment of the present application, as shown in fig. 2, the ADAS scenario includes at least one sub-application scenario, and the acquiring a positive sample data set and a negative sample data set in the ADAS scenario to serve as a first training set includes steps S201 to S202 as follows:
step S201, a positive sample data set, a negative sample data set and a false alarm image data set under each sub-application scene are collected.
In specific implementation, because the ADAS scene is widely applied, the construction of a data set covering the whole is the premise of realizing target detection optimization and effectively inhibiting false alarms. In the ADAS scene, the effective data may be affected by the illumination, weather, road environment, or even the style of a building, so the embodiment of the present application sub-classifies various types of data scenes according to the requirement of false alarm suppression, and further performs data balance between different application scenes, so as to cover the generation sources of various types of false alarms as much as possible.
In specific implementation, a positive sample data set, a negative sample data set and a false alarm image data set under each sub-scene need to be acquired. The false alarm image data set is added in the training set and the verification set independently, namely, the part of data is added as a class of the data set, and the class of data is kept in the same order of magnitude as that of single-class positive sample data.
Step S202, classifying the positive sample data set, the negative sample data set and the false alarm image data set in each sub-application scene according to preset classification items to obtain a training set in each sub-application scene, wherein the preset classification items include but are not limited to time, weather, climate, road environment and image quality.
In specific implementation, after the data sets in each sub-application scene are collected, the data in the data sets need to be classified according to preset classification items, where the preset classification items may include time (day, night, dawn, dusk, etc.), weather (sunny, cloudy, rainy, fog, snow, etc.), climate (dry (autumn, winter/frigid zone/northern/western), wet (spring/temperate zone/middle), hot (summer/subtropical zone/tropical/south), etc.), road environment (urban arterial road, urban alley, underground parking lot, expressway, urban rural road, tunnel lamp), image quality (clear, fuzzy, lightproof, etc.), and the like. The specific scene classification manner is not limited to the above-mentioned cases, and those skilled in the art can flexibly set the scene classification manner according to the actual situation, which is not listed herein. Through the process, data balance among different specific application scenes can be realized so as to cover various false alarm generation sources as far as possible.
As a preferred implementation manner of the embodiment of the present application, as shown in fig. 3, the sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain the second training set includes steps S301 to S303 as follows:
step S301, a negative sample region clipping queue is constructed, and the negative sample region clipping queue is updated in real time.
In specific implementation, the data increment strategy for virtual suppression proposed in the embodiment of the present application first needs to construct a negative sample region clipping queue roi based on the acquired negative sample data setFPI.e. picture initialized to all 0, and updating the queue in real time, as shown in the following formula (1):
roiFP＝{roi0}∪{a|aconf＞Thresh，atype＝FP}， (1)
wherein α is the total negative examples marked off from the anchor point in the training.
Step S302, sample the negative sample region clipping queue, and randomly fill the sampled negative sample region into the positive sample region in the positive sample data set.
In specific implementation, the real-time updated negative sample region clipping queue is sampled, and the sampled negative sample region is randomly filled into a real positive sample region.
Step S303, randomly removing the positive sample labeling box corresponding to the positive sample region to obtain the second training set.
In specific implementation, after the sampled negative sample region is randomly filled into the real positive sample region, the labeling frame of the positive sample is randomly removed to serve as a new training set.
As a preferred implementation of the embodiment of the present application, the preset loss function is formed by the following formula (2):
wherein the content of the first and second substances,
i is the current iteration number, ImaxAs a result of the total number of iterations,
in order to be an illustrative function of the system,
is LdetThe complement of the class prediction, where,
where N is the number of anchor point predictions as positive samples, LconfAs a function of classification loss, LlocEquation (3) is a general form of the loss function, and it can be found that the loss value of this training is 0 when there is no positive sample.
The embodiment of the application is based on a preset loss function (formula (2)) obtained after a general loss function is improved, and aims to emphasize the value of a false alarm in training, in an initial training stage, the iteration times are not sufficient, the target detection capability of a DCNN model is weak, the distinguishing of positive and negative samples is not significant enough, along with the increase of the iteration times, the model has strong enough distinguishing capability on positive and negative samples of a general scene, the proportion of the negative samples does not need to be defined by relying on the detection of the positive samples, the classification loss of the positive and negative samples is taken into a total loss function to be beneficial to the inhibition of the false alarm, and harmonic parameters are set, and along with the deepening of the iteration times, the proportion of the negative samples taken into the loss function can be gradually increased.
As a preferred implementation manner of the embodiment of the present application, as shown in fig. 4, after the optimizing the target detection model based on the preset loss function to perform the false alarm suppression of target detection according to the optimized target detection model, the method includes steps S401 to S403 as follows:
step S401, sampling the first training set and the second training set to obtain a sampling data set.
Due to the complexity of the ADAS scene, the target detection algorithm cannot achieve perfect beauty, and continuous iteration updating must be performed by combining a real application scene. Therefore, the embodiment of the application provides a test and application driven data completion strategy for false alarm suppression. In specific implementation, an original training set, that is, a training set obtained by mixing the first training set and the second training set, is sampled to obtain a sampled data set.
And S402, mixing the sampling data set and the negative sampling data set according to a preset proportion to obtain a mixed data set.
In specific implementation, the negative sample data set in the embodiment of the present application is acquired through the following processes: (1) the method comprises the following steps of collecting and returning images in an actual application scene, wherein the collection strategy mainly comprises the following steps: detecting dense scenes, high-speed driving scenes of vehicles, abnormal driving behavior scenes (such as sudden braking) and false detection and false alarm conditions of test and user feedback through an algorithm; (2) and reconstructing a verification set and a test set, and analyzing a negative sample scene. Specifically, false detection and false alarm analysis are carried out on collected and returned data, and specific false detection and false alarm scene examples are collected; (3) and reproducing the image scene subjected to false detection in a targeted manner and acquiring a specific negative sample data set.
And then, mixing the acquired negative sample data set and the sampling data set according to a certain proportion to obtain a mixed data set.
Step S403, fine-tuning the target detection model according to the mixed data set.
In specific implementation, the data set is input into a target detection model for training, so that the DCNN target detection model is finely adjusted according to a training result.
From the above description, it can be seen that the present invention achieves the following technical effects: through the measures of reasonably constructing a data set, optimizing a data increment method, a target detection algorithm loss function and the like, the target under the ADAS scene is reasonably planned in positive and negative sample increment and sampling proportion, end-to-end optimization is realized in the training process of the target detection algorithm based on the DCNN, and false alarms are effectively inhibited on the basis of keeping detection precision and even slightly improving the detection precision.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer-executable instructions and that, although a logical order is illustrated in the flowcharts, in some cases, the steps illustrated or described may be performed in an order different than presented herein.
According to an embodiment of the present invention, there is also provided an apparatus for implementing the above target detection false alarm suppression method for an ADAS scene, as shown in fig. 5, the apparatus includes: the system comprises an acquisition module 1, an increment module 2, a training module 3 and an optimization module 4. The acquisition module 1 in the embodiment of the application is used for acquiring a positive sample data set and a negative sample data set in an ADAS scene to serve as a first training set; the increment module 2 in the embodiment of the application is configured to sample the negative sample data set and randomly fill the negative sample data set into the positive sample data set to obtain a second training set; the training module 3 of the embodiment of the application is configured to input the first training set and the second training set into a preset network model for training to obtain a target detection model; the optimization module 4 of the embodiment of the application is configured to optimize the target detection model based on a preset loss function, so as to perform false alarm suppression of target detection according to the optimized target detection model.
As a preferred implementation manner of the embodiment of the present application, the ADAS scenario includes at least one sub-application scenario, and the acquisition module includes: the acquisition unit is used for acquiring a positive sample data set, a negative sample data set and a false alarm image data set in each sub-application scene; and the classification unit is used for classifying the positive sample data set, the negative sample data set and the false alarm image data set in each sub-application scene according to preset classification items to obtain a training set in each sub-application scene, wherein the preset classification items include but are not limited to time, weather, climate, road environment and image quality.
As a preferred implementation manner of the embodiment of the present application, the increment module includes: the construction unit is used for constructing a negative sample region clipping queue and updating the negative sample region clipping queue in real time; the sampling unit is used for sampling the negative sample region clipping queue and randomly filling the sampled negative sample region into the positive sample region in the positive sample data set; and the removing unit is used for randomly removing the positive sample marking frame corresponding to the positive sample region to obtain the second training set.
As a preferred implementation of the embodiment of the present application, the apparatus further includes: a sampling module, configured to sample the first training set and the second training set to obtain a sampled data set; the mixing module is used for mixing the sampling data set and the negative sampling data set according to a preset proportion to obtain a mixed data set; and the fine tuning module is used for fine tuning the target detection model according to the mixed data set.
For the specific connection relationship between the modules and the units and the functions performed, please refer to the detailed description of the method, which is not repeated herein.
According to an embodiment of the present invention, there is also provided a computer apparatus including: one or more processors; storage means for storing one or more programs; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method as previously described.
There is also provided, in accordance with an embodiment of the present invention, a computer-readable storage medium having stored thereon computer instructions, which when executed by a processor, implement the steps of the method as previously described.
As shown in fig. 6, the electronic device includes one or more processors 31 and a memory 32, and one processor 31 is taken as an example in fig. 6.
The control unit may further include: an input device 33 and an output device 34.
The processor 31, the memory 32, the input device 33 and the output device 34 may be connected by a bus or other means, and fig. 6 illustrates the connection by a bus as an example.
The processor 31 may be a Central Processing Unit (CPU). The Processor 31 may also be other general purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components, or combinations thereof. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory 32, 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. The processor 31 executes various functional applications of the server and data processing by running non-transitory software programs, instructions and modules stored in the memory 32, namely, implements the target detection false alarm suppression method for ADAS scenarios of the above-described method embodiments.
The memory 32 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 according to use of a processing device operated by the server, and the like. Further, the memory 32 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 32 may optionally include memory located remotely from the processor 31, which may be connected to a network connection device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input device 33 may receive input numeric or character information and generate key signal inputs related to user settings and function control of the processing device of the server. The output device 34 may include a display device such as a display screen.
One or more modules are stored in the memory 32, which when executed by the one or more processors 31 perform the methods as previously described.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The computer instructions are for causing the computer to perform the above-described target detection false alarm suppression method for an ADAS scenario.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, the principle and the implementation of the present invention are explained by applying the specific embodiments in the present invention, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (10)
1. A target detection false alarm suppression method for an ADAS scene is characterized by comprising the following steps:
collecting a positive sample data set and a negative sample data set in an ADAS scene to serve as a first training set;
sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain a second training set;
inputting the first training set and the second training set into a preset network model for training to obtain a target detection model;
and optimizing the target detection model based on a preset loss function so as to carry out false alarm suppression of target detection according to the optimized target detection model.
2. The method of claim 1, wherein the ADAS scenario includes at least one sub-application scenario, and the collecting a positive sample data set and a negative sample data set in the ADAS scenario as the first training set comprises:
acquiring a positive sample data set, a negative sample data set and a false alarm image data set in each sub-application scene;
classifying the positive sample data set, the negative sample data set and the false alarm image data set in each sub-application scene according to preset classification items to obtain a training set in each sub-application scene, wherein the preset classification items include but are not limited to time, weather, climate, road environment and image quality.
3. The method of claim 1, wherein the sampling and randomly populating the negative sample data set into the positive sample data set to obtain a second training set comprises:
constructing a negative sample region clipping queue, and updating the negative sample region clipping queue in real time;
sampling the negative sample region clipping queue, and randomly filling the sampled negative sample region into a positive sample region in the positive sample data set;
and randomly removing the positive sample marking frame corresponding to the positive sample region to obtain the second training set.
4. The method of claim 1, wherein the predetermined loss function is comprised of the following equation:
wherein the content of the first and second substances,
i is the current iteration number, ImaxAs a result of the total number of iterations,
in order to be an illustrative function of the system,
is LdetThe complement of the class prediction is selected,
wherein the content of the first and second substances,
n is the number of anchor point predictions as positive samples, LconfAs a function of classification loss, LlocFor bounding box loss function, α is the reconciliation parameter between the classification and the bounding box.
5. The method of claim 1, wherein the optimizing the target detection model based on a predetermined loss function to perform target detection false alarm suppression according to the optimized target detection model comprises:
sampling the first training set and the second training set to obtain a sampled data set;
mixing the sampling data set and the negative sample data set according to a preset proportion to obtain a mixed data set;
and fine-tuning the target detection model according to the mixed data set.
6. An object detection false alarm suppression apparatus for an ADAS scenario, comprising:
the acquisition module is used for acquiring a positive sample data set and a negative sample data set in an ADAS scene to serve as a first training set;
the increment module is used for sampling the negative sample data set and randomly filling the negative sample data set into the positive sample data set to obtain a second training set;
the training module is used for inputting the first training set and the second training set into a preset network model for training so as to obtain a target detection model;
and the optimization module is used for optimizing the target detection model based on a preset loss function so as to carry out false alarm suppression of target detection according to the optimized target detection model.
7. The apparatus of claim 6, wherein the ADAS scene comprises at least one sub-application scene, the acquisition module comprises:
the acquisition unit is used for acquiring a positive sample data set, a negative sample data set and a false alarm image data set in each sub-application scene;
and the classification unit is used for classifying the positive sample data set, the negative sample data set and the false alarm image data set in each sub-application scene according to preset classification items to obtain a training set in each sub-application scene, wherein the preset classification items include but are not limited to time, weather, climate, road environment and image quality.
8. The apparatus of claim 6, wherein the increment module comprises:
the construction unit is used for constructing a negative sample region clipping queue and updating the negative sample region clipping queue in real time;
the sampling unit is used for sampling the negative sample region clipping queue and randomly filling the sampled negative sample region into the positive sample region in the positive sample data set;
and the removing unit is used for randomly removing the positive sample marking frame corresponding to the positive sample region to obtain the second training set.
9. An electronic device, comprising:
one or more processors;
storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-5.
10. A non-transitory readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010095000.XA CN111310660B (en) | 2020-02-14 | 2020-02-14 | Target detection false alarm suppression method and device for ADAS scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010095000.XA CN111310660B (en) | 2020-02-14 | 2020-02-14 | Target detection false alarm suppression method and device for ADAS scene |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111310660A true CN111310660A (en) | 2020-06-19 |
| CN111310660B CN111310660B (en) | 2024-05-17 |
Family
ID=71161726
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010095000.XA Active CN111310660B (en) | 2020-02-14 | 2020-02-14 | Target detection false alarm suppression method and device for ADAS scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111310660B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112085056A (en) * | 2020-08-05 | 2020-12-15 | 深圳市优必选科技股份有限公司 | Target detection model generation method, device, equipment and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109614967A (en) * | 2018-10-10 | 2019-04-12 | 浙江大学 | A kind of detection method of license plate based on negative sample data value resampling |
| CN109784190A (en) * | 2018-12-19 | 2019-05-21 | 华东理工大学 | A kind of automatic Pilot scene common-denominator target Detection and Extraction method based on deep learning |
| WO2019127924A1 (en) * | 2017-12-29 | 2019-07-04 | 深圳云天励飞技术有限公司 | Sample weight allocation method, model training method, electronic device, and storage medium |
| CN109977812A (en) * | 2019-03-12 | 2019-07-05 | 南京邮电大学 | A kind of Vehicular video object detection method based on deep learning |
| CN110020664A (en) * | 2019-01-31 | 2019-07-16 | 浙江工业大学 | A kind of positive negative sample balance method of deep learning target detection |
| CN110321928A (en) * | 2019-06-03 | 2019-10-11 | 深圳中兴网信科技有限公司 | Generation method, computer equipment and the readable storage medium storing program for executing of environment measuring model |
| CN110659545A (en) * | 2018-06-29 | 2020-01-07 | 比亚迪股份有限公司 | Training method of vehicle recognition model, vehicle recognition method and device and vehicle |
-
2020
- 2020-02-14 CN CN202010095000.XA patent/CN111310660B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019127924A1 (en) * | 2017-12-29 | 2019-07-04 | 深圳云天励飞技术有限公司 | Sample weight allocation method, model training method, electronic device, and storage medium |
| CN110659545A (en) * | 2018-06-29 | 2020-01-07 | 比亚迪股份有限公司 | Training method of vehicle recognition model, vehicle recognition method and device and vehicle |
| CN109614967A (en) * | 2018-10-10 | 2019-04-12 | 浙江大学 | A kind of detection method of license plate based on negative sample data value resampling |
| CN109784190A (en) * | 2018-12-19 | 2019-05-21 | 华东理工大学 | A kind of automatic Pilot scene common-denominator target Detection and Extraction method based on deep learning |
| CN110020664A (en) * | 2019-01-31 | 2019-07-16 | 浙江工业大学 | A kind of positive negative sample balance method of deep learning target detection |
| CN109977812A (en) * | 2019-03-12 | 2019-07-05 | 南京邮电大学 | A kind of Vehicular video object detection method based on deep learning |
| CN110321928A (en) * | 2019-06-03 | 2019-10-11 | 深圳中兴网信科技有限公司 | Generation method, computer equipment and the readable storage medium storing program for executing of environment measuring model |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112085056A (en) * | 2020-08-05 | 2020-12-15 | 深圳市优必选科技股份有限公司 | Target detection model generation method, device, equipment and storage medium |
| CN112085056B (en) * | 2020-08-05 | 2023-12-29 | 深圳市优必选科技股份有限公司 | Target detection model generation method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111310660B (en) | 2024-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111626208B (en) | Method and device for detecting small objects | |
| US11380104B2 (en) | Method and device for detecting illegal parking, and electronic device | |
| CN111680377B (en) | Traffic situation simulation method, system and electronic equipment | |
| Hmida et al. | Hardware implementation and validation of a traffic road sign detection and identification system | |
| CN112307978A (en) | Target detection method and device, electronic equipment and readable storage medium | |
| CN114492022A (en) | Road condition sensing data processing method, device, equipment, program and storage medium | |
| CN113657299A (en) | Traffic accident determination method and electronic equipment | |
| CN114495060B (en) | Road traffic marking recognition method and device | |
| CN111142402A (en) | Simulation scene construction method and device and terminal | |
| CN115761668A (en) | Camera stain recognition method and device, vehicle and storage medium | |
| CN111310660B (en) | Target detection false alarm suppression method and device for ADAS scene | |
| CN113160272B (en) | Target tracking method and device, electronic equipment and storage medium | |
| CN109800684A (en) | The determination method and device of object in a kind of video | |
| CN112765302A (en) | Method and device for processing position information and computer readable medium | |
| CN117197796A (en) | Vehicle shielding recognition method and related device | |
| Adam et al. | Robustness and deployability of deep object detectors in autonomous driving | |
| CN116977484A (en) | Image desensitizing method, device, electronic equipment and storage medium | |
| CN115797880A (en) | Method and device for determining driving behavior, storage medium and electronic device | |
| CN115439815A (en) | Driving condition identification method, device, equipment, medium and vehicle | |
| CN114419018A (en) | Image sampling method, system, device and medium | |
| CN113609980A (en) | Lane line sensing method and device for automatic driving vehicle | |
| CN114077797A (en) | Automatic driving test scene design method and device based on road traffic regulations | |
| CN110659384B (en) | Video structured analysis method and device | |
| WO2024066798A1 (en) | Vehicle control method and apparatus, and device and storage medium | |
| CN118193366A (en) | Test case generation method and device, electronic equipment and storage medium |
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 |