CN116434197A - Abnormal action detection closed loop iterative optimization method, system, medium and equipment - Google Patents
Abnormal action detection closed loop iterative optimization method, system, medium and equipment Download PDFInfo
- Publication number
- CN116434197A CN116434197A CN202111663629.0A CN202111663629A CN116434197A CN 116434197 A CN116434197 A CN 116434197A CN 202111663629 A CN202111663629 A CN 202111663629A CN 116434197 A CN116434197 A CN 116434197A
- Authority
- CN
- China
- Prior art keywords
- data
- abnormal
- model
- normal
- probability
- 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.)
- Pending
Links
Images
Classifications
-
- 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
- Traffic Control Systems (AREA)
Abstract
The application discloses an abnormal action detection closed loop iteration optimization method, system, medium and equipment, and belongs to the technical field of data optimization. The method comprises the following steps: in the vehicle-mounted terminal, the vehicle-mounted model carries out coarse screening on collected face video data of a driver to obtain massive low-precision video data; classifying mass low-precision video data in a cloud server to obtain first normal data, first abnormal data and in-doubt data; accurately classifying the in-doubt data into second normal data and second abnormal data; and performing model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle-mounted model and an optimized judgment model, and updating. According to the data collection method and device, the data are collected in a data collection mode with high recall rate and low precision, the collected data are processed by using the model on the cloud server, the precision of data processing is improved, and the data classification is more and more accurate.
Description
Technical Field
The present disclosure relates to the field of data optimization technologies, and in particular, to a closed loop iterative optimization method, system, medium, and apparatus for detecting abnormal actions.
Background
During driving of a vehicle, the driver often presents some dangerous driving actions or behaviors. For example, smoking, drinking water, making a yawning, making a call, closing eyes, etc. during driving. Therefore, accurate detection of driver behavior is of great importance for safe driving. The conventional method in the prior art is that a low-recall high-precision data collection mode is adopted at a vehicle end to collect driver data, and after the vehicle end finishes data collection and classification processing of driver behaviors, the data which cannot be analyzed and accurately analyzed out of the low-recall high-precision data are reconfirmed in a manual labeling mode. The high-precision data collection mode with low recall rate is adopted, so that the precision of the collected data is high, and the process of manual reconfirmation is redundant; in addition, the data quantity of the collected data is small through a low-recall-rate high-precision data collection mode, a lot of data containing various behavior information is filtered out through the low-recall-rate high-precision data collection mode, so that when the model is further optimized, the model cannot learn more contents, and the improvement space of the model is small.
Disclosure of Invention
Aiming at the problems that in the prior art, when data collection and processing of driver behaviors are carried out, a high-precision low-recall-rate data collection mode is adopted in a vehicle-mounted terminal, so that data containing a plurality of information are filtered, the available data size is small, and the model cannot be further optimized by utilizing the data, the abnormal action detection closed-loop iteration optimization method, system, medium and equipment are provided.
In one technical scheme of the present application, an abnormal motion detection closed loop iterative optimization method is provided, including: in the vehicle-mounted terminal, the vehicle-mounted model carries out coarse screening on collected face video data of a driver to obtain massive low-precision video data; classifying mass low-precision video data in a cloud server through a judgment model to obtain first normal data, first abnormal data and in-doubt data; accurately classifying the in-doubt data, and classifying the in-doubt data into second normal data and second abnormal data; and carrying out model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle-mounted model and an optimized judgment model, and respectively updating the vehicle-mounted model and the judgment model.
Optionally, the vehicle-mounted model performs coarse screening on the collected face video data of the driver to obtain massive low-precision video data, including: analyzing the face data by using the vehicle-mounted model, and determining a first probability that the face data belongs to normal data and a second probability that the face data belongs to abnormal data; if the first probability or the second probability is larger than the corresponding first preset threshold, the face data are used as massive low-precision video data and uploaded to the cloud server.
Optionally, classifying, in the cloud server, the massive low-precision video data through the judgment model to obtain first normal data, first abnormal data and in-doubt data, including: carrying out sliding window processing on massive low-precision video data by using a large model to obtain a plurality of window data; analyzing each window data, and respectively determining a third probability of the window data belonging to normal data and a fourth probability of the window data belonging to abnormal data; and respectively calculating the average value of a plurality of third probabilities and fourth probabilities, if the third probability average value is larger than a corresponding second preset threshold value, determining the mass low-precision video data as first normal data or first abnormal data, otherwise, determining the mass low-precision video data as suspicious data.
Optionally, the method includes classifying, in the cloud server, the massive low-precision video data through a judgment model to obtain first normal data, first abnormal data and in-doubt data, and further includes: analyzing the plurality of window data by utilizing a time sequence model, and determining a fifth probability that massive low-precision video data belong to normal data or abnormal data in a time dimension; judging the massive low-precision video data according to the fifth probability, the third probability average value and the fourth probability average value, and determining the massive low-precision video data as first normal data, first abnormal data or in-doubt data.
Optionally, the method includes classifying, in the cloud server, the massive low-precision video data through a judgment model to obtain first normal data, first abnormal data and in-doubt data, and further includes: detecting the gesture of a driver in the massive low-precision video data by using an auxiliary detection model to obtain judgment scores corresponding to the massive low-precision video data; and judging the mass low-precision video data according to the judgment score, the fifth probability, the third probability average value and the fourth probability average value, and determining that the screened video data is first normal data, first abnormal data or in-doubt data.
Optionally, performing model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle model and an optimized judgment model, including: and carrying out model training by utilizing massive first normal data, first abnormal data, second normal data and second abnormal data, and respectively optimizing the optimized vehicle-mounted model and the optimized judgment model.
Optionally, the method further comprises: and performing data evaluation by using the first normal data, the first abnormal data, the second normal data and the second abnormal data, so that a comprehensive evaluation result of driving of the driver can be obtained.
In one technical solution of the present application, there is provided an abnormal motion detection closed loop iterative optimization system, including: the vehicle-mounted terminal performs coarse screening on the collected face video data of the driver by using a vehicle-mounted model to obtain massive low-precision video data; the cloud server classifies the massive low-precision video data through the judgment model to obtain first normal data, first abnormal data and in-doubt data, accurately classifies the in-doubt data, and classifies the in-doubt data into second normal data and second abnormal data; and the post-processing module is used for carrying out model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle-mounted model and an optimized judgment model, and respectively updating the vehicle-mounted model and the judgment model.
In one aspect of the present application, a computer-readable storage medium is provided, the storage medium storing computer instructions that are operative to perform the method of aspect one.
In one aspect of the present application, there is provided a computer device comprising a processor and a memory, the memory storing computer instructions, wherein: the processor operates on computer instructions to perform the method of scheme one.
The beneficial effects of this application are: according to the data collection method, the data are collected in a high-recall-rate low-precision data collection mode, a large amount of sufficient data containing driver behavior information are obtained, the collected data are processed by the aid of the model on the cloud server, the data processing precision is improved, the data are optimized, the optimized model can accurately process the data, and the data are more and more accurately classified.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief description will be given below of the drawings that are needed in the embodiments or the prior art descriptions, it being obvious that the drawings in the following description are some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow diagram of one embodiment of a closed loop iterative optimization method for abnormal action detection of the present application;
FIG. 2 is a schematic diagram of one embodiment of an abnormal-motion detection closed-loop iterative optimization system of the present application;
FIG. 3 is a schematic diagram of an example of an abnormal-action detection closed-loop iterative optimization system of the present application.
Specific embodiments thereof have been shown by way of example in the drawings and will herein be described in more detail. These drawings and the written description are not intended to limit the scope of the inventive concepts in any way, but to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of 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 apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
The terms "first," "second," "third," "fourth" and the like in the description and in the claims of this application and in the above-described figures, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that embodiments of the present application described herein may be capable of operation in sequences other than those illustrated or described herein, for example. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a product or apparatus that comprises a sequence of steps or elements is not necessarily limited to those elements that are expressly listed or inherent to such product or apparatus, but may include other elements not expressly listed or inherent to such product or apparatus.
During driving of a vehicle, the driver often presents some dangerous driving actions or behaviors. For example, smoking, drinking water, making a yawning, making a call, closing eyes, etc. during driving. Therefore, accurate detection of driver behavior is of great importance for safe driving. The conventional method in the prior art is that a low-recall high-precision data collection mode is adopted at a vehicle end to collect driver data, and after the vehicle end finishes data collection and classification processing of driver behaviors, the data which cannot be analyzed and accurately analyzed out of the low-recall high-precision data are reconfirmed in a manual labeling mode. The high-precision data collection mode with low recall rate is adopted, so that the precision of the collected data is high, and the process of manual reconfirmation is redundant; in addition, the data quantity of the collected data is small through a low-recall-rate high-precision data collection mode, a lot of data containing various behavior information is filtered out through the low-recall-rate high-precision data collection mode, so that when the model is further optimized, the model cannot learn more contents, and the improvement space of the model is small.
Aiming at the problems, the application provides a closed loop iterative optimization method, a closed loop iterative optimization system, a closed loop iterative optimization medium and closed loop iterative optimization equipment for abnormal action detection. The method comprises the following steps: in the vehicle-mounted terminal, the vehicle-mounted model carries out coarse screening on collected face video data of a driver to obtain massive low-precision video data; classifying mass low-precision video data in a cloud server through a judgment model to obtain first normal data, first abnormal data and in-doubt data; accurately classifying the in-doubt data, and classifying the in-doubt data into second normal data and second abnormal data; and carrying out model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle-mounted model and an optimized judgment model, and respectively updating the vehicle-mounted model and the judgment model.
According to the abnormal action detection closed loop iterative optimization method, the data is collected in a high-recall-rate low-precision data collection mode, a large amount of sufficient data containing driver behavior information is obtained, the collected data is processed by using the model on the cloud server, the accuracy of data processing is improved, the model is optimized by using the data, the optimized model can accurately process the data, and the classification of the data is more accurate.
The following describes the technical solutions of the present application and how the technical solutions of the present application solve the above technical problems in detail with specific embodiments. The following embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. Embodiments of the present application will be described below with reference to the accompanying drawings.
FIG. 1 illustrates one embodiment of a closed loop iterative optimization method for abnormal action detection of the present application.
In the embodiment shown in fig. 1, the abnormal motion detection closed loop iterative optimization method of the present application includes a process S101, in which a vehicle-mounted model performs coarse screening on collected face video data of a driver to obtain massive low-precision video data.
In this embodiment, the abnormal actions of the driver aimed at in the present application mainly include smoking, drinking water, making a yawning, closing eyes, making a call, and the like. When the data are collected, the vehicle-mounted camera can be used for shooting the driver, so that the behavior data of the driver can be obtained. And then analyzing the data shot by the camera by using the vehicle-mounted model, and judging whether the driver shot by the camera has the abnormal actions. When specific analysis is carried out, pictures shot by a camera can be cut to obtain face data of a driver, the data are analyzed and judged to obtain probability that the data belong to abnormal data, coarse screening is carried out by using probability values, and finally massive low-precision video data are obtained.
Optionally, the vehicle-mounted model performs coarse screening on the collected face video data of the driver to obtain massive low-precision video data, including: analyzing the face data by using the vehicle-mounted model, and determining a first probability that the face data belongs to normal data and a second probability that the face data belongs to abnormal data; if the first probability or the second probability is larger than the corresponding first preset threshold, the face data are used as massive low-precision video data and uploaded to the cloud server.
In this alternative embodiment, the face data is analyzed using an on-board model to determine a first probability that the face video data belongs to normal data and a second probability that the face video data belongs to abnormal data. I.e. analyzing whether the driver has smoking, drinking, etc. and determining the corresponding first probability and second probability. And comparing the relation between the first probability and the second probability with a corresponding first preset threshold value, and if the first probability or the second probability is larger than the first preset threshold value, taking the face video data as low-precision video data and uploading the low-precision video data to a server. The method and the device adopt a data collection mode with low precision and high recall rate, and the first preset threshold value is generally low in value so as to obtain more data. It should be noted that, the value of the first preset threshold value can be reasonably set according to actual requirements, and the application is not particularly limited.
Specifically, for example, when the face data of the driver in three frames of images are analyzed, the probability of the first frame belonging to the normal data is 40%, the probability of the second frame belonging to the abnormal data is 50%, the probability of the third frame data belonging to the normal data is 20%, and the probability of the third frame data belonging to the abnormal data is 22%. Assuming that the value of the first preset threshold is 30%, judging that the face video data corresponding to the first frame picture and the face video data corresponding to the second frame picture can be uploaded to the cloud server as low-precision video data, and filtering the face video data corresponding to the third frame picture. Compared with the mode of high-precision low recall in the prior art, a higher threshold is often set, and if the first preset threshold is set to 45%, only the face video data corresponding to the second frame of picture can be uploaded to the cloud server as low-precision video data. The data collection mode in the prior art causes little data volume, and difficult recall of difficult data, and is inconvenient for subsequent operation and model optimization.
In the embodiment shown in fig. 1, the abnormal motion detection closed loop iterative optimization method of the present application includes a process S102, in which massive low-precision video data is classified by a judgment model in a cloud server, so as to obtain first normal data, first abnormal data and in-doubt data.
In this embodiment, the cloud server may carry a judgment model with stronger processing capability to further analyze and classify the massive low-precision video data, and subdivide the massive low-precision video data into normal data, first abnormal data and suspicious data. The method comprises the steps that in-doubt data are mass low-precision video data, after judgment of a model, whether the low-precision video data are normal data or abnormal data cannot be clearly indicated, and therefore the data serve as in-doubt data.
Optionally, the cloud server classifies the massive low-precision video data through a judgment model to obtain first normal data, first abnormal data and in-doubt data, including: carrying out sliding window processing on massive low-precision video data by using a large model to obtain a plurality of window data; analyzing each window data, and respectively determining a third probability of the window data belonging to normal data and a fourth probability of the window data belonging to abnormal data; and respectively calculating the average value of the third probabilities and the fourth probabilities, if the third probability average value or the fourth probability average value is larger than a corresponding second preset threshold value, determining the mass low-precision video data as first normal data or first abnormal data, otherwise, determining the mass low-precision video data as suspicious data.
In this alternative embodiment, the screened video data is further analyzed in the cloud server by a large model with very high processing power. Firstly, sliding window processing is carried out on the screening video data to obtain a plurality of window data. And respectively analyzing the plurality of window data, and respectively determining the third probability that each window data belongs to normal data and the fourth probability that each window data belongs to abnormal data. Calculating the average value of a plurality of third probabilities and fourth probabilities, judging the relation between the third probability average value or the fourth probability average value and a second preset threshold value, if the third probability average value is larger than the corresponding second preset threshold value, determining the low-precision video number as first normal data, if the fourth probability average value is larger than the corresponding second preset threshold value, determining the low-precision video data as first abnormal data, otherwise, determining the low-precision video data as doubtful data.
Specifically, the screening video data may be 5 seconds of video, and the plurality of window data obtained after the sliding window processing may be 2 seconds of video data. The second probability of the window data is obtained after analysis of the 2 second window data. A second probability example of window data is shown therein as follows:
| window data 1 | Window data 2 | Window data 3 | Probability mean value | |
| First normal data | 55％ | 60％ | 65％ | 60％ |
| First abnormal data | 30％ | 15％ | 25* | 35％ |
As shown in the table above, in the obtaining of the probability average, the third probability average belonging to the first normal data is 60%, the fourth probability average belonging to the first abnormal data is 35%, and if the second preset threshold is set to 50%, it can be determined that the screened video data belongs to the first normal data. It should be noted that the above arrangements are merely illustrative of the principles of the present application, and specific values thereof may be reasonably selected according to actual requirements.
Optionally, the method includes classifying, in the cloud server, the massive low-precision video data through a judgment model to obtain first normal data, first abnormal data and in-doubt data, and further includes: analyzing the plurality of window data by utilizing a time sequence model, and determining a fifth probability that massive low-precision video data belong to normal data or abnormal data in a time dimension; judging the massive low-precision video data according to the fifth probability, the third probability average value and the fourth probability average value, and determining the massive low-precision video data as first normal data, first abnormal data or in-doubt data.
In this optional embodiment, in order to improve the accuracy of data analysis, the present application uses a time sequence model to analyze multiple window data in the cloud server, and in a time dimension, the fifth probability that the massive low-accuracy video data belongs to normal data or abnormal data. And then comprehensively analyzing the fifth probability and the third probability average value and the fourth probability average value analyzed by the large model to determine that the low-precision video data is first normal data, first abnormal data or in-doubt data.
Specifically, when the fifth probability, the third probability average value and the fourth probability average value are comprehensively judged, different weights can be allocated for comprehensive consideration, or other modes can be adopted. And the final determined result is more accurate.
Optionally, the method includes classifying, in the cloud server, the massive low-precision video data through a judgment model to obtain first normal data, first abnormal data and in-doubt data, and further includes: detecting the gesture of a driver in the massive low-precision video data by using an auxiliary detection model to obtain judgment scores corresponding to the massive low-precision video data; and judging the mass low-precision video data according to the judgment score, the fifth probability, the third probability average value and the fourth probability average value, and determining that the screened video data is first normal data, first abnormal data or in-doubt data.
In this embodiment, in an actual scenario, the driver may have some low-head activities such as playing a cell phone. At this time, the eyes of the driver are blocked, and since the eyes are not detected in the process of analyzing and judging by the model, an erroneous judgment result may occur. Therefore, aiming at the special situation, the application detects the gesture of the driver in the screening video data by using the auxiliary detection model to acquire the judgment score corresponding to the screening video data. When the auxiliary detection model specifically judges the posture of the driver, the rotation angle of the head of the driver, the distance between the upper eyelid and the lower eyelid of the eye and the like are comprehensively judged, and the judgment scores of the driver behavior belonging to normal actions or abnormal actions are respectively obtained. And comprehensively judging the mass low-precision video data by the judgment score, the fifth probability, the third probability average value and the fourth probability average value, and finally determining the low-precision video data as first normal data, first abnormal data or in-doubt data, thereby improving the detection accuracy.
In the embodiment shown in fig. 1, the abnormal motion detection closed loop iterative optimization method of the present application includes a process S103, which accurately classifies the in-doubt data into the second normal data and the second abnormal data.
In this embodiment, the suspicious data in the mass low-precision video data that cannot be classified by the model is handled by a manual labeling method, and the suspicious data is classified into the second normal data and the second abnormal data.
In the embodiment shown in fig. 1, the abnormal motion detection closed loop iterative optimization method of the present application includes a process S104, performing model training by using first normal data, first abnormal data, second normal data and second abnormal data, to obtain an optimized vehicle model and an optimized judgment model, and updating the vehicle model and the judgment model respectively.
Optionally, performing model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle model and an optimized judgment model, including: and carrying out model training by utilizing massive first normal data, first abnormal data, second normal data and second abnormal data, and respectively carrying out optimization determination on the optimized vehicle-mounted model and the optimized judgment model.
In this alternative embodiment, a large number of first normal data, first abnormal data, second normal data, and second abnormal data, which are accurately classified, are obtained through the above-described process. By training the model by using the accurate data, the model trained by using the data can be better fitted under the condition that the data is better, so that the accuracy of the model in processing the data is higher.
Optionally, the abnormal action detection closed loop iterative optimization method of the present application further includes: and performing data evaluation by using the first normal data, the first abnormal data, the second normal data and the second abnormal data, so that a comprehensive evaluation result of driving of the driver can be obtained.
In this alternative embodiment, the obtained first normal data, first abnormal data, second normal data, and second abnormal data are also used for evaluating data. The data collection mode with low precision and high recall rate is adopted, so that the data volume is greatly improved, various situations can be perfectly covered, and the evaluation process is more comprehensive.
According to the abnormal action detection closed loop iterative optimization method, the data is collected in a high-recall-rate low-precision data collection mode, massive low-precision video data containing driver behavior information are obtained fully, the collected data are processed by using the model on the cloud server, the precision of data processing is improved, the data are optimized, the optimized model can process the data accurately, and the classification of the data is more accurate. And data collection is performed with low precision through high recall rate, and data processing is performed with high precision through high recall rate, so that data with higher availability is finally obtained. The model is trained and optimized, so that a closed loop is formed in the whole process, the optimization is continuously carried out, collected data meets requirements more and more, the data is cleaner and more optimized, the data which cannot be detected and judged before the model is optimized, the model can be completely used for judging after the optimization, the manual labeling process is finally omitted, and the data processing capacity of the model is improved.
FIG. 2 illustrates one embodiment of a closed loop iterative optimization system for abnormal action detection of the present application.
In the embodiment shown in fig. 2, the abnormal-action detection closed-loop iterative optimization system of the present application includes: the vehicle-mounted terminal 201 performs coarse screening on the collected face video data of the driver by using a vehicle-mounted model to obtain massive low-precision video data; the cloud server 202 classifies the massive low-precision video data through a judgment model to obtain first normal data, first abnormal data and in-doubt data, accurately classifies the in-doubt data, and classifies the in-doubt data into second normal data and second abnormal data; the post-processing module 203 performs model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle model and an optimized judgment model, and updates the vehicle model and the judgment model respectively.
Optionally, in the vehicle-mounted terminal 201, the face data is analyzed by using a vehicle-mounted model, and a first probability that the face data belongs to normal data and a second probability that the face data belongs to abnormal data are determined; if the first probability or the second probability is larger than the corresponding first preset threshold, the face data are used as massive low-precision video data and uploaded to the cloud server.
Optionally, in the cloud server 202, sliding window processing is performed on the massive low-precision video data by using a large model to obtain a plurality of window data; analyzing each window data, and respectively determining a third probability of the window data belonging to normal data and a fourth probability of the window data belonging to abnormal data; and respectively calculating the average value of a plurality of third probabilities and fourth probabilities, if the third probability average value is larger than a corresponding second preset threshold value, determining the mass low-precision video data as first normal data or first abnormal data, otherwise, determining the mass low-precision video data as suspicious data.
Optionally, in the cloud server 202, analyzing the plurality of window data by using a time sequence model, and determining a fifth probability that the massive low-precision video data belongs to normal data or abnormal data in a time dimension; judging the massive low-precision video data according to the fifth probability, the third probability average value and the fourth probability average value, and determining the massive low-precision video data as first normal data, first abnormal data or the number of doubts.
Optionally, in the cloud server 202, the gesture of the driver in the massive low-precision video data is detected by using an auxiliary detection model, so as to obtain a judgment score corresponding to the massive low-precision video data; and judging the mass low-precision video data according to the judgment score, the fifth probability, the third probability average value and the fourth probability average value, and determining that the screened video data is first normal data, first abnormal data or the number of doubts.
FIG. 3 illustrates one example of a closed loop iterative optimization system for abnormal action detection of the present application.
As shown in fig. 3, first, each vehicle-mounted terminal collects data in a high-recall low-precision manner, and then transmits the data to a cloud server. And on the cloud server, processing data by using the large model, the sequence model and the auxiliary detection model, and judging the data which cannot be judged by some models by a manual labeling method. And the final data are evaluated and the model is trained, so that the model is more and more optimized, the data which cannot be detected and judged before can be judged through the model completely after optimization, and finally, the manual labeling process is omitted.
According to the abnormal action detection closed loop iteration optimization system, the data is collected in a data collection mode with high recall rate and low precision, a large amount of sufficient data containing driver behavior information is obtained, the collected data is processed by using the model on the cloud server, the accuracy of data processing is improved, the model is optimized by using the data, the optimized model can accurately process the data, and the classification of the data is more accurate.
In one embodiment of the present application, a computer readable storage medium stores computer instructions operable to perform the abnormal-action detection closed-loop iterative optimization method described in any of the embodiments. Wherein the storage medium may be directly in hardware, in a software module executed by a processor, or in a combination of the two.
A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium.
The processor may be a central processing unit (English: central Processing Unit; CPU; for short), or other general purpose processor, digital signal processor (English: digital Signal Processor; for short DSP), application specific integrated circuit (English: application Specific Integrated Circuit; ASIC; for short), field programmable gate array (English: field Programmable Gate Array; FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof, etc. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In the alternative, the processor and the storage medium may reside as discrete components in a user terminal.
In one embodiment of the present application, a computer device includes a processor and a memory storing computer instructions, wherein: the processor operates the computer instructions to perform the abnormal action detection closed loop iterative optimization method described in any of the embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of elements is merely a logical functional division, and there may be additional divisions of actual implementation, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
The foregoing is only examples of the present application, and is not intended to limit the scope of the patent application, and all equivalent structural changes made by the specification and drawings of the present application, or direct or indirect application in other related technical fields, are included in the scope of the patent protection of the present application.
Claims (10)
1. An abnormal motion detection closed loop iterative optimization method is characterized by comprising the following steps:
in the vehicle-mounted terminal, the vehicle-mounted model carries out coarse screening on collected face video data of a driver to obtain massive low-precision video data;
classifying the mass low-precision video data in a cloud server through a judgment model to obtain first normal data, first abnormal data and in-doubt data;
accurately classifying the suspicious data, and classifying the suspicious data into second normal data and second abnormal data;
and performing model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle-mounted model and an optimized judgment model, and updating the vehicle-mounted model and the judgment model respectively.
2. The abnormal motion detection closed loop iterative optimization method according to claim 1, wherein the vehicle model performs coarse screening on the collected driver face video data to obtain massive low-precision video data, and the method comprises the following steps:
analyzing the face data by using the vehicle-mounted model, and determining a first probability that the face data belongs to normal data and a second probability that the face data belongs to abnormal data;
and if the first probability or the second probability is larger than a corresponding first preset threshold, the face data are used as the massive low-precision video data and uploaded to the cloud server.
3. The abnormal motion detection closed loop iterative optimization method according to claim 1, wherein the classifying the massive low-precision video data in the cloud server through a judgment model to obtain first normal data, first abnormal data and in-doubt data comprises:
carrying out sliding window processing on the massive low-precision video data by using a large model to obtain a plurality of window data;
analyzing each window data, and respectively determining a third probability of the window data belonging to normal data and a fourth probability of the window data belonging to abnormal data;
and respectively calculating the average values of the third probabilities and the fourth probabilities, if the third probability average value is larger than the fourth probability average value by a corresponding second preset threshold value, determining the mass low-precision video data as the first normal data or the first abnormal data, otherwise, determining the mass low-precision video data as the suspicious data.
4. The abnormal motion detection closed loop iterative optimization method according to claim 3, wherein the classifying the massive low-precision video data in the cloud server through a judgment model to obtain first normal data, first abnormal data and in-doubt data, further comprises:
analyzing a plurality of window data by using a time sequence model, and determining a fifth probability that the massive low-precision video data belong to normal data or abnormal data in a time dimension;
judging the massive low-precision video data according to the fifth probability, the third probability average value and the fourth probability average value, and determining the massive low-precision video data as the first normal data, the first abnormal data or the suspicious data.
5. The abnormal motion detection closed loop iterative optimization method according to claim 4, wherein the classifying the massive low-precision video data in the cloud server through a judgment model to obtain first normal data, first abnormal data and in-doubt data, further comprises:
detecting the gesture of a driver in the massive low-precision video data by using an auxiliary detection model to obtain a judgment score corresponding to the massive low-precision video data;
and judging the mass low-precision video data according to the judgment score, the fifth probability, the third probability average value and the fourth probability average value, and determining the screened video data as the first normal data, the first abnormal data or the in-doubt data.
6. The abnormal motion detection closed loop iterative optimization method according to claim 1, wherein the performing model training by using the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle model and an optimized judgment model comprises:
and performing model training by using massive first normal data, first abnormal data, second normal data and second abnormal data, and respectively optimizing the optimized vehicle-mounted model and the optimized judgment model.
7. The abnormal-motion detection closed-loop iterative optimization method according to claim 1, further comprising:
and performing data evaluation by using the first normal data, the first abnormal data, the second normal data and the second abnormal data, so that a comprehensive evaluation result of driving of the driver can be obtained.
8. An abnormal-motion detection closed-loop iterative optimization system, comprising:
the vehicle-mounted terminal performs coarse screening on the collected face video data of the driver by using a vehicle-mounted model to obtain massive low-precision video data;
the cloud server classifies the massive low-precision video data through a judgment model to obtain first normal data, first abnormal data and in-doubt data, accurately classifies the in-doubt data, and classifies the in-doubt data into second normal data and second abnormal data;
and the post-processing module is used for carrying out model training by utilizing the first normal data, the first abnormal data, the second normal data and the second abnormal data to obtain an optimized vehicle-mounted model and an optimized judgment model, and respectively updating the vehicle-mounted model and the judgment model.
9. A computer readable storage medium having stored thereon computer instructions operative to perform the abnormal-action detection closed-loop iterative optimization method of any one of claims 1-7.
10. A computer device comprising a processor and a memory, the memory storing computer instructions, wherein: the processor operates the computer instructions to perform the abnormal-action detection closed-loop iterative optimization method of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111663629.0A CN116434197A (en) | 2021-12-31 | 2021-12-31 | Abnormal action detection closed loop iterative optimization method, system, medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111663629.0A CN116434197A (en) | 2021-12-31 | 2021-12-31 | Abnormal action detection closed loop iterative optimization method, system, medium and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116434197A true CN116434197A (en) | 2023-07-14 |
Family
ID=87091231
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111663629.0A Pending CN116434197A (en) | 2021-12-31 | 2021-12-31 | Abnormal action detection closed loop iterative optimization method, system, medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116434197A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116665025A (en) * | 2023-07-31 | 2023-08-29 | 福思(杭州)智能科技有限公司 | Data closed-loop method and system |
-
2021
- 2021-12-31 CN CN202111663629.0A patent/CN116434197A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116665025A (en) * | 2023-07-31 | 2023-08-29 | 福思(杭州)智能科技有限公司 | Data closed-loop method and system |
| CN116665025B (en) * | 2023-07-31 | 2023-11-14 | 福思(杭州)智能科技有限公司 | Data closed-loop method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110807385B (en) | Target detection method, target detection device, electronic equipment and storage medium | |
| CN110020592B (en) | Object detection model training method, device, computer equipment and storage medium | |
| CN112132119B (en) | Passenger flow statistical method and device, electronic equipment and storage medium | |
| WO2019051941A1 (en) | Method, apparatus and device for identifying vehicle type, and computer-readable storage medium | |
| CN108875600A (en) | A kind of information of vehicles detection and tracking method, apparatus and computer storage medium based on YOLO | |
| CN111291817A (en) | Image recognition method and device, electronic equipment and computer readable medium | |
| CN113723157B (en) | Crop disease identification method and device, electronic equipment and storage medium | |
| CN106056083A (en) | Information processing method and terminal | |
| CN116434197A (en) | Abnormal action detection closed loop iterative optimization method, system, medium and equipment | |
| CN115205615A (en) | Vehicle cleanliness detection method and device and computer readable storage medium | |
| US20220405527A1 (en) | Target Detection Methods, Apparatuses, Electronic Devices and Computer-Readable Storage Media | |
| CN116704490B (en) | License plate recognition method, license plate recognition device and computer equipment | |
| US20170309040A1 (en) | Method and device for positioning human eyes | |
| US11887292B1 (en) | Two-step anti-fraud vehicle insurance image collecting and quality testing method, system and device | |
| CN116840835B (en) | Fall detection method, system and equipment based on millimeter wave radar | |
| CN106251364A (en) | Method for tracking target and device | |
| CN114495244B (en) | Electrographic calculation data anomaly detection method, device, equipment and storage medium | |
| CN110543827A (en) | multi-class vehicle detection method based on Gaussian mixture model and deep learning | |
| CN116052231A (en) | Palm vein recognition method, device, equipment and storage medium | |
| CN109543610A (en) | Vehicle detecting and tracking method, device, equipment and storage medium | |
| CN111553408B (en) | Automatic test method for video recognition software | |
| CN111860081B (en) | Time sequence signal classification method and device and electronic equipment | |
| CN116363542A (en) | Off-duty event detection method, apparatus, device and computer readable storage medium | |
| WO2022263904A1 (en) | Target detection methods, apparatuses, electronic devices and computer-readable storage media | |
| CN112580462A (en) | Feature point selection method, terminal 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 |