Driver posture regression method based on general probability distribution

Abstract

The present invention provides a driver posture regression method based on general probability distribution, comprising: obtaining a driver image; inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver posture; the probability prediction network is obtained based on training of a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with the set general probability distribution; according to the probability distribution function, the driver posture prediction value and confidence are calculated. On the one hand, the present invention enables the driver posture regression result to include a prediction value and a corresponding probability, thereby providing a basis for subsequent application to set the weight of the prediction value according to the probability; on the other hand, it can obtain a more real and reliable prediction value for the objective uneven distribution of the driver posture.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a driver posture regression method based on general probability distribution. Background Art

[0002] With the development of the economy and society, motor vehicles have become one of the main modes of transportation. At the same time, the number of traffic accidents has inevitably increased. Drivers' irregular driving behaviors, such as fatigue driving, smoking, and making handheld phone calls, are one of the major hidden dangers of road safety today. Therefore, in order to reduce the risk of road accidents, it has become a consensus in the industry to monitor drivers' driving behaviors.

[0003] In driver behavior monitoring, driver posture detection is an important part. The prior art has proposed some driver posture regression methods based on deep learning, the principle of which is to train a deep learning network model based on driver pictures or video samples and driver posture labels (i.e., driver posture true values), and use the trained model to regress the driver posture in the input picture / video.

[0004] Although the above-mentioned existing technologies can solve the problem of driver posture regression to a certain extent, considering that the driver posture results are often used for further driver behavior judgment, a single output driver posture often cannot guarantee accuracy, which may have an adverse impact on subsequent driver behavior judgment tasks. Therefore, it can be considered that the existing technologies have the defects of high false alarm risk and poor controllability.

[0005] Therefore, how to provide a driver posture regression method based on probability (i.e., the uncertainty of the driver posture obtained by regression) distribution has become an urgent problem to be solved in the industry. Summary of the invention

[0006] The present invention provides a driver posture regression method based on general probability distribution, which is used to solve the defects of high false alarm risk and poor controllability in the prior art and realize probability-based driver posture regression.

[0007] The present invention provides a driver posture regression method based on general probability distribution, comprising:

[0008] Acquire the driver image;

[0009] The driver image is input into a probability prediction network to obtain a probability distribution function of the driver's posture; the probability prediction network is trained based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with a set general probability distribution;

[0010] According to the probability distribution function, the driver posture prediction value and confidence level are calculated.

[0011] According to a driver posture regression method based on general probability distribution provided by the present invention, the variables of the training loss function of the probability prediction network include the probability true value of the discrete posture point;

[0012] The discrete posture point is a driver posture quantization value obtained by discretizing the driver posture in a set method within the quantization interval of the driver posture;

[0013] If the driver posture quantization value of the label is equal to any discrete posture point, then the probability truth value satisfies:

[0014] P′(C)=1

[0015] If the driver posture quantization value of the label is not equal to any discrete posture point, then the probability truth value satisfies:

[0016]

[0017]

[0018] Wherein, C is the quantized value of the driver's posture of the label; a is the maximum discrete posture point less than C; b is the minimum discrete posture point greater than C; P′(C) is the probability truth value of the label at C; P′(a) is the probability truth value of the label at a; P′(b) is the probability truth value of the label at b.

[0019] According to a driver posture regression method based on general probability distribution provided by the present invention, the step of calculating the driver posture prediction value and confidence level according to the probability distribution function comprises:

[0020] The driver's posture prediction value is calculated according to the first formula:

[0021]

[0022] In the formula, is the predicted value of the driver's posture; i is the i-th discrete posture interval; P(y i ) is the probability distribution function, representing the probability of the i-th discrete posture interval; n is the number of discrete posture intervals;

[0023] The discrete posture intervals refer to a plurality of intervals separated by discrete posture points as boundaries within the quantization interval of the driver's posture; the discrete posture points refer to the driver's posture quantization values ​​obtained by discretization in a set method within the quantization interval of the driver's posture;

[0024] The confidence level is calculated according to the second formula:

[0025]

[0026] Where α is the predicted value of the driver’s posture confidence level.

[0027] According to a driver posture regression method based on general probability distribution provided by the present invention, the loss function includes a first term and a second term; the weight of the first term is P′(a); the weight of the second term is P′(b).

[0028] According to a driver posture regression method based on general probability distribution provided by the present invention, the first term and / or the second term of the loss function is an L1 loss function.

[0029] According to a driver posture regression method based on general probability distribution provided by the present invention, the first term and / or the second term of the loss function is a cross entropy loss function.

[0030] The present invention also provides a driver posture regression system based on general probability distribution, comprising:

[0031] An acquisition module, used for acquiring a driver image;

[0032] A computing module, used for inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver's posture; the probability prediction network is obtained by training based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with a set general probability distribution;

[0033] The prediction module is used to calculate the driver posture prediction value and confidence level according to the probability distribution function.

[0034] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned driver posture regression methods based on general probability distribution are implemented.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described driver posture regression methods based on general probability distribution.

[0036] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any of the above-mentioned driver posture regression methods based on general probability distribution are implemented.

[0037] The driver posture regression method based on general probability distribution provided by the present invention predicts the driver posture through a probability prediction network trained with general probability distribution labels, so that the probability distribution of the driver posture is consistent with the general probability distribution of the labels. Therefore, on the one hand, the driver posture regression result includes a predicted value and a corresponding probability, thereby providing a basis for setting the weight of the predicted value according to the probability in subsequent applications; on the other hand, it can obtain a more realistic and reliable predicted value for the objective uneven distribution of the driver posture. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 It is a flow chart of the driver posture regression method based on general probability distribution provided by the present invention;

[0040] Figure 2 is a first schematic diagram of a general probability distribution provided by an embodiment of the present invention;

[0041] Figure 3 is a second schematic diagram of a general probability distribution provided by an embodiment of the present invention;

[0042] Figure 4 is a third schematic diagram of a general probability distribution provided by an embodiment of the present invention;

[0043] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Combine the following Figure 1-Figure 4 The driver posture regression method based on general probability distribution of the present invention is described.

[0046] like Figure 1 As shown, an embodiment of the present invention provides a driver posture regression method based on general probability distribution, comprising:

[0047] Step 101, obtaining a driver image;

[0048] Step 103, inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver's posture; the probability prediction network is trained based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with the set general probability distribution;

[0049] Step 105: Calculate the driver posture prediction value and confidence level according to the probability distribution function.

[0050] In step 101, the driver image may be independent image data or any frame in the input driver video.

[0051] In this embodiment, the driver's posture may be the posture angle of the driver's head, that is, the pitch angle, roll angle, and yaw angle of the driver's head, or the posture angle of the driver's line of sight.

[0052] The general probability distribution is set based on the actual distribution probability of the driver's posture, which is different from the common, fixed Gaussian distribution in other fields.

[0053] For example, due to the limitations of the human musculoskeletal structure, the pitch angle of the driver's head posture angle can only vary within a range of about plus or minus 60 degrees, and the yaw angle can only vary within a range of about plus or minus 90 degrees. If we refer to the common Gaussian distribution in other fields for probability prediction, taking the yaw angle prediction value of -30 degrees as an example, the probability distribution function should obtain an extreme value at -30 degrees, and the probability value (i.e., confidence) decreases symmetrically on both sides of -30 degrees.

[0054] However, this probability distribution is obviously inconsistent with the actual situation. The yaw angle of the driver's head is still considered. In actual driving, the probability of the driver's head facing forward (yaw angle of about 0 degrees) and facing the left, middle and right rearview mirrors (yaw angles of about -15 degrees, 15 degrees and 50 degrees) is significantly greater than other possible yaw angles. In addition, the probability of the yaw angle corresponding to the driver's gaze at the center console, the mobile phone, etc. should also reach a maximum value.

[0055] Therefore, simply giving a probability distribution function of the driver's posture based on Gaussian distribution still cannot accurately predict the driver's posture. The prior art neither provides a technical revelation on the probability value (i.e., confidence) of the driver's posture regression result, nor provides a technical revelation that the probability distribution should consider the actual situation of the driver's posture. Therefore, the discovery of the above technical problem itself is also one of the creative manifestations of this embodiment.

[0056] The beneficial effects of this embodiment are:

[0057] The driver's posture is predicted by a probability prediction network trained with general probability distribution labels, so that the probability distribution of the driver's posture is consistent with the general probability distribution of the label. On the one hand, the driver's posture regression result includes the predicted value and the corresponding probability, which provides a basis for setting the weight of the predicted value according to the probability in subsequent applications; on the other hand, it can derive a more realistic and reliable prediction value based on the objective uneven distribution of the driver's posture.

[0058] According to the above embodiment, in this embodiment:

[0059] The variables of the training loss function of the probability prediction network include the true probability values ​​of discrete posture points;

[0060] The discrete posture point is a driver posture quantization value obtained by discretizing the driver posture in a set method within the quantization interval of the driver posture;

[0061] If the driver posture quantization value of the label is equal to any discrete posture point, then the probability truth value satisfies:

[0062] P′(C)=1

[0063] If the driver posture quantization value of the label is not equal to any discrete posture point, then the probability truth value satisfies:

[0064]

[0065]

[0066] Wherein, C is the quantized value of the driver's posture of the label; a is the maximum discrete posture point less than C; b is the minimum discrete posture point greater than C; P′(C) is the probability truth value of the label at C; P′(a) is the probability truth value of the label at a; P′(b) is the probability truth value of the label at b.

[0067] The step of calculating the driver posture prediction value and confidence level according to the probability distribution function comprises:

[0068] The driver's posture prediction value is calculated according to the first formula:

[0069]

[0070] In the formula, is the predicted value of the driver's posture; i is the i-th discrete posture interval; P(y i ) is the probability distribution function, representing the probability of the i-th discrete posture interval; n is the number of discrete posture intervals;

[0071] The discrete posture intervals refer to a plurality of intervals separated by discrete posture points as boundaries within the quantization interval of the driver's posture; the discrete posture points refer to the driver's posture quantization values ​​obtained by discretization in a set method within the quantization interval of the driver's posture;

[0072] The confidence level is calculated according to the second formula:

[0073]

[0074] Where α is the predicted value of the driver’s posture confidence level.

[0075] The loss function includes a first term and a second term; the weight of the first term is P′(a); the weight of the second term is P′(b).

[0076] The first term and / or the second term of the loss function is an L1 loss function; or, the first term and / or the second term of the loss function is a cross entropy loss function.

[0077] This embodiment provides a method for calculating the predicted value and the confidence level. Figures 2 to 4 , and further explain and illustrate the predicted value and confidence.

[0078] If the probability distribution function of the probability prediction network output is Figure 2 If the above formula is consistent, then the predicted value should be at the short vertical line in the figure, and its corresponding confidence (that is, the probability given by the probability distribution function at the predicted value) is low. The reason is that Figure 2 The general probability distribution given has two peaks, and the calculated prediction value lies between the two peaks. From a probabilistic point of view, the uncertainty of this prediction value is high.

[0079] If the probability distribution function of the probability prediction network output is Figure 2 or Figure 3 The above formula shows that the predicted value should be at the short vertical line in the figure. The confidence of the two is relatively high because Figure 2 , Figure 3 The predicted values ​​do not belong to the valley area of ​​these two general probability distributions.

[0080] This embodiment also explains and illustrates the training process of the probability prediction network. In order to simplify the amount of calculation required for the training process and improve the accuracy of regression, the loss function in this embodiment can be based on discrete posture points, that is, the quantization interval of the driver's posture (if the driver's posture regression is used to regress the driver's head pitch angle, the quantization value is the pitch angle) is divided into multiple discrete posture intervals through each discrete posture point, and training is performed based on the discrete posture intervals. The discrete posture interval in which the predicted value is located is determined, and the loss function is set according to the two endpoints of the discrete posture interval in which the true value (i.e., the label) is located. This can effectively reduce the amount of calculation and the time required for the training process while ensuring the training effect.

[0081] The beneficial effects of this embodiment are:

[0082] 1. It solves the single problem of Gaussian distribution assumption of uncertainty and is more universal;

[0083] 2. Improve the accuracy of model regression tasks and obtain the top-k regression distribution map to facilitate model optimization.

[0084] The driver posture regression device based on general probability distribution provided by the present invention is described below. The driver posture regression device based on general probability distribution described below and the driver posture regression method based on general probability distribution described above can refer to each other.

[0085] The embodiment of the present invention further provides a driver posture regression system based on general probability distribution, comprising:

[0086] An acquisition module, used for acquiring a driver image;

[0087] A computing module, used for inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver's posture; the probability prediction network is obtained by training based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with a set general probability distribution;

[0088] The prediction module is used to calculate the driver posture prediction value and confidence level according to the probability distribution function.

[0089] The variables of the training loss function of the probability prediction network include the true probability values ​​of discrete posture points;

[0090] The discrete posture point is a driver posture quantization value obtained by discretizing the driver posture in a set method within the quantization interval of the driver posture;

[0091] If the driver posture quantization value of the label is equal to any discrete posture point, then the probability truth value satisfies:

[0092] P′(C)=1

[0093] If the driver posture quantization value of the label is not equal to any discrete posture point, then the probability truth value satisfies:

[0094]

[0095]

[0096] Wherein, C is the quantized value of the driver's posture of the label; a is the maximum discrete posture point less than C; b is the minimum discrete posture point greater than C; P′(C) is the probability truth value of the label at C; P′(a) is the probability truth value of the label at a; P′(b) is the probability truth value of the label at b.

[0097] The step of calculating the driver posture prediction value and confidence level according to the probability distribution function comprises:

[0098] The driver's posture prediction value is calculated according to the first formula:

[0099]

[0100] In the formula, is the predicted value of the driver's posture; i is the i-th discrete posture interval; P(y i ) is the probability distribution function, representing the probability of the i-th discrete posture interval; n is the number of discrete posture intervals;

[0101] The discrete posture intervals refer to a plurality of intervals separated by discrete posture points as boundaries within the quantization interval of the driver's posture; the discrete posture points refer to the driver's posture quantization values ​​obtained by discretization in a set method within the quantization interval of the driver's posture;

[0102] The confidence level is calculated according to the second formula:

[0103]

[0104] Where α is the predicted value of the driver’s posture confidence level.

[0105] The loss function includes a first term and a second term; the weight of the first term is P′(a); the weight of the second term is P′(b).

[0106] The first term and / or the second term of the loss function is an L1 loss function; or, the first term and / or the second term of the loss function is a cross entropy loss function.

[0107] The beneficial effects of this embodiment are:

[0108] The driver's posture is predicted by a probability prediction network trained with general probability distribution labels, so that the probability distribution of the driver's posture is consistent with the general probability distribution of the label. On the one hand, the driver's posture regression result includes the predicted value and the corresponding probability, which provides a basis for setting the weight of the predicted value according to the probability in subsequent applications; on the other hand, it can derive a more realistic and reliable prediction value based on the objective uneven distribution of the driver's posture.

[0109] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a driver posture regression method based on a general probability distribution, the method comprising: obtaining a driver image; inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver posture; the probability prediction network is obtained by training based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with the set general probability distribution; according to the probability distribution function, the driver posture prediction value and confidence are calculated.

[0110] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0111] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the driver posture regression method based on general probability distribution provided by the above methods, and the method includes: obtaining a driver image; inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver posture; the probability prediction network is trained based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with the set general probability distribution; according to the probability distribution function, the driver posture prediction value and confidence are calculated.

[0112] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the driver posture regression method based on general probability distribution provided by the above-mentioned methods, the method comprising: obtaining a driver image; inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver posture; the probability prediction network is trained based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with the set general probability distribution; according to the probability distribution function, the driver posture prediction value and confidence are calculated.

[0113] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0114] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A driver posture regression method based on general probability distribution, characterized in that: include: Acquire the driver image; Inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver's posture; The probability prediction network is obtained by training based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with the set general probability distribution; According to the probability distribution function, a driver posture prediction value and a confidence level are calculated; The variables of the training loss function of the probability prediction network include the true probability values ​​of discrete posture points; The discrete posture point is a driver posture quantization value obtained by discretizing the driver posture in a set method within the quantization interval of the driver posture; If the driver posture quantization value of the label is equal to any discrete posture point, then the probability truth value satisfies: ； If the driver posture quantization value of the label is not equal to any discrete posture point, then the probability truth value satisfies: ； ； In the formula, A driver posture quantized value for the tag; is less than The maximum discrete posture point of is greater than The minimum discrete posture point of For the label The probability of the true value at ; For the label The probability of the true value at ; For the label The probability of the true value.

2. The driver posture regression method based on general probability distribution according to claim 1, characterized in that: The step of calculating the driver posture prediction value and confidence level according to the probability distribution function comprises: The driver's posture prediction value is calculated according to the first formula: ； In the formula, is the driver posture prediction value; For the Discrete posture intervals; is the probability distribution function, indicating the The probability of a discrete posture interval; is the number of discrete posture intervals; The discrete posture intervals refer to a plurality of intervals separated by discrete posture points as boundaries within the quantization interval of the driver's posture; the discrete posture points refer to the driver's posture quantization values ​​obtained by discretization in a set method within the quantization interval of the driver's posture; The confidence level is calculated according to the second formula: ； In the formula, The predicted value of the driver's posture confidence level.

3. The driver posture regression method based on general probability distribution according to claim 1, characterized in that: The loss function includes a first term and a second term; the weight of the first term is ; The weight of the second item is .

4. The driver posture regression method based on general probability distribution according to claim 3 is characterized in that: The first term and / or the second term of the loss function is an L1 loss function.

5. The driver posture regression method based on general probability distribution according to claim 3 is characterized in that: The first term and / or the second term of the loss function is a cross entropy loss function.

6. A driver posture regression system based on general probability distribution, characterized in that: include: An acquisition module, used for acquiring a driver image; A calculation module, used for inputting the driver image into a probability prediction network to obtain a probability distribution function of the driver's posture; The probability prediction network is obtained by training based on a training set; the training set includes one-to-one corresponding samples and labels, and the probability distribution of the labels is consistent with the set general probability distribution; A prediction module, used to calculate the driver posture prediction value and confidence level according to the probability distribution function; The variables of the training loss function of the probability prediction network include the true probability values ​​of discrete posture points; The discrete posture point is a driver posture quantization value obtained by discretizing the driver posture in a set method within the quantization interval of the driver posture; If the driver posture quantization value of the label is equal to any discrete posture point, then the probability truth value satisfies: ； If the driver posture quantization value of the label is not equal to any discrete posture point, then the probability truth value satisfies: ； ； In the formula, A driver posture quantized value for the tag; is less than The maximum discrete posture point of is greater than The minimum discrete posture point of For the label The probability of the true value at ; For the label The probability of the true value at ; For the label The probability of the true value.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the driver posture regression system method based on general probability distribution as described in any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the driver posture regression system method based on general probability distribution as described in any one of claims 1 to 5 are implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the driver posture regression system method based on general probability distribution as described in any one of claims 1 to 5 are implemented.

Images