Intelligent safety air bag ignition system and method based on passenger classification and identification
By combining image processing and sensor technology, identifying occupant types and optimizing airbag ignition algorithms, the problem of poor occupant type identification in the prior art is solved, and the protection effect of different types of occupants is improved, especially in infants and children protection. In terms of this, the risk of accidental injury is reduced.
Patent Information
- Application Number
- CN202411114385.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-19
AI Technical Summary
The prior art in occupant type recognition, especially for infants and young children under the age of 12, has poor effect, resulting in the active safety system's effectiveness in protecting these people is lower than that of car safety seats, and may even cause damage, and image recognition The accuracy is greatly affected by the external environment and sample quality.
Design an intelligent airbag system based on occupant recognition, combining image processing technology, image recognition technology and external sensors, identify occupant types through data acquisition module, image processing module and classifier module, and obtain collision data through acceleration sensors, optimized Airbag ignition algorithm to improve protection effect.
Effectively identify infants, children, adult women and adult men during the applicable period of safety seats, and predict occupants' height through sensor information, improve the accuracy of occupants type identification and the protection effect of airbag ignition, and reduce the risk of ignition by mistake.
Smart Images

Figure CN118977671A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of intelligent automobiles, and in particular relates to an intelligent safety airbag ignition system and algorithm based on occupant classification and recognition. Background Art
[0002] In today's vehicle active safety field, especially in the vehicle's intelligent restraint system, the core concept is that the car actively makes corresponding adaptive adjustments for different occupant types. When a vehicle collides with a vehicle that may cause casualties, the restraint systems such as airbags and seat belts in the vehicle will protect the occupants from severe impact and collision, and avoid secondary collision injuries.
[0003] Effective identification of occupant types has become an important prerequisite for achieving active safety. At present, the identification of occupant types often uses image recognition methods based on machine learning. Machine learning specifically studies how computers simulate or realize human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. It is the core of artificial intelligence and the fundamental way to make computers intelligent. This is an effective tool, but it also has its limitations. The quality of samples, the influence of the external environment, the occupant's own actions, etc., will affect the accuracy of image recognition to a large extent. Especially for infants and young children under the age of 12 who are in the applicable period of car safety seats, the effect of image recognition is often very poor. The protection effect of active safety for this group of people is often lower than that of the car safety seat itself, and may even have a counter-effect on them and cause damage. At this time, the execution of some active safety systems including airbags should be suppressed. Therefore, it is of great significance to supervise the results of image recognition, complement and verify the results of sensor recognition and image recognition, so as to improve the accuracy and reliability of occupant type recognition.
[0004] After the occupant type is identified, the traditional airbag moving window ignition method also needs to be changed accordingly, so as to improve the protective effect of the airbag on the occupant through occupant type identification. Summary of the invention
[0005] The purpose of the present invention is to design an intelligent airbag system based on occupant identification. The system includes an occupant type identification module and an airbag ignition module. The occupant type identification module uses image processing technology, image recognition technology and external sensors to obtain occupant type information, effectively identify infants, children, adult women, 50th percentile adult males and 95th percentile adult males in the applicable period of the safety seat, and predict the occupant height through sensor information to realize the supervision of the identification result. The airbag ignition module mainly improves the mobile window ignition method to improve the protection effect of different types of occupants.
[0006] To achieve the above object, the present invention adopts the following technical solutions:
[0007] An intelligent safety airbag ignition system based on occupant classification recognition, characterized in that it includes an occupant type recognition module and an airbag ignition execution module, wherein the occupant type recognition module includes a data acquisition module, an image processing module, and a classifier module;
[0008] The data acquisition module includes a pressure sensor, a seat belt buckle sensor, an ultrasonic sensor, an image sensor and an acceleration sensor. There are two pressure sensors, which are respectively placed on the car seat and the car backrest to collect the body pressure of the occupant. The seat belt buckle sensor is placed at the seat belt buckle to collect the seat belt buckle status. The ultrasonic sensor is placed inside the roof just above the occupant to collect the distance between the occupant's head and the roof. The image sensor is a three-channel vehicle-mounted camera, which is placed above the windshield to collect the upper body image information of the passenger in the front passenger seat. The acceleration sensor is placed on the B-pillar of the car to collect the acceleration of the car.
[0009] The image processing module provides known samples and unknown samples for the learning and training of the classifier module by intercepting the region of interest, graying, filtering and smoothing, edge detection and feature extraction;
[0010] The classifier module classifier is used to classify the occupants in terms of gender and age;
[0011] The airbag ignition execution module is used to ignite the safety airbag.
[0012] An intelligent airbag ignition method based on occupant classification recognition comprises the following steps:
[0013] S1. Determine the occupant type through the data acquisition module, the image processing module, and the classifier module, and calculate the ignition threshold A;
[0014] S2. Obtain the acceleration data of the vehicle through the acceleration sensor, where t1 is the collision start time, t2 is the best ignition time obtained after optimizing the image measurement results, and t3 = t2 + 30ms is the time when the airbag is fully filled with gas;
[0015] S3, detect and calculate the integral result S according to the time window width;
[0016] S4. When the integral result S is greater than or equal to the ignition threshold A, the airbag is ignited and the actual ignition time is recorded;
[0017] When the integration result S is less than the ignition threshold A, the airbag does not ignite, and the process returns to step S2.
[0018] Preferably, the step S1 specifically includes the following steps:
[0019] S1.1:
[0020] The system starts to operate when the pressure sensor placed on the car seat is affected by a change in pressure;
[0021] S1.2:
[0022] During the operation of the system, if the pressure sensor at the seat back receives a pressure signal, the seat belt buckle sensor will detect whether the seat belt buckle is closed;
[0023] If the seat belt buckle is closed, the seat belt closing period obtained by the seat belt buckle sensor is used as the occupant's sitting posture correction time T;
[0024] When the passenger sits in a correct posture, the ultrasonic sensor placed on the top of the passenger works to measure the distance between the top of the passenger's head and the roof during the T period. The distances corresponding to different times are sent to the on-board computer, and the minimum distance from the top of the head to the roof is selected. This distance value is set as x1 in centimeters, and the time corresponding to x1 is set as t.
[0025] The distance from the roof to the seat cushion is set as x2, in centimeters;
[0026] Then the height l from the buttocks to the top of the head is x2-x1, in centimeters;
[0027] The average height of adult males / the average height from hips to head of adult males sitting upright is a;
[0028] The predicted height of the occupant is L = a*l, in centimeters;
[0029] S1.3:
[0030] The electronic signal obtained by the pressure sensor at the seat cushion is sent to the on-board computer. If the computer recognizes that the body pressure distribution image is a seamless regular pattern, it means that the safety seat, not the passenger, is in contact with the seat cushion. The pending result 1 is obtained, and the passenger on the seat cushion is an infant in the applicable period of the safety seat.
[0031] S1.4:
[0032] When the occupant on the seat cushion is not an infant within the applicable period of the safety seat, a picture of the occupant at a corresponding frame number at time t is obtained through the vehicle-mounted camera, and the picture is cropped;
[0033] S1.5:
[0034] Use the trained classifier to detect the gender and age of the occupants on the processed image samples;
[0035] If the occupant is detected to be younger than 12 years old, the pending result 1 is changed to the real result 1, and the occupant is an infant within the applicable period of the safety seat;
[0036] If the age of the occupant is greater than 12 years old but less than 18 years old, the result is 2, and the occupant is a child;
[0037] If the age is greater than 18 years old and the deep neural network classifier determines that the occupant's gender is female, the result 3 is obtained, and the occupant is an adult female;
[0038] If the age is greater than 18 years old and the gender is male, the predicted height L of the occupant obtained in S3.2 is used for judgment;
[0039] If the occupant's predicted height L is greater than 188 cm, the result is 4, and the occupant is a 95th percentile male;
[0040] If the occupant's predicted height L is less than 188 cm, the result is 5, and the occupant is a 50th percentile male;
[0041] S1.6:
[0042] The acceleration data of the car is obtained through the acceleration sensor, where t1 is the start time of the collision, t2 is the best ignition time obtained after optimizing the image measurement results, and t3=t2+30ms is the time when the airbag is fully filled with gas.
[0043] Preferably, in step S3, the window widths are selected as 20ms, 30ms, 40ms, 50ms, 60ms, and 70ms respectively by enumeration method, and the formula Divide by the time window width w to get A(t,w), normalize A(t,w), and divide A(t,w) by the maximum value A(t,w) max , and obtain the normalized index
[0044]
[0045] Technical effects and advantages of the present invention:
[0046] 1) The occupant type recognition module of the present invention utilizes image processing technology, image recognition technology and external sensors to obtain occupant type information, effectively identifies infants, children, adult females, 50th percentile adult males and 95th percentile adult males within the applicable period of the safety seat, and predicts the occupant's height through sensor information to supervise the recognition results.
[0047] 2) The airbag ignition module of the present invention is mainly an improvement on the moving window ignition algorithm to enhance the protection effect for different types of occupants. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1This is a flowchart of occupant type identification of the patented invention.
[0049] Figure 2 This is a schematic diagram of the sensor positions used in the present invention.
[0050] Figure 3 It is the result of processing the collision test data with a collision speed of 50km / h according to the normalized index.
[0051] Figure 4 This is the flow chart of the patented airbag ignition algorithm of this invention. DETAILED DESCRIPTION
[0052] The following will be combined with the attached embodiment of the present invention Figure 1-Figure 4 , the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0053] Example 1
[0054] like Figure 1 As shown, an intelligent safety airbag ignition system based on occupant classification recognition includes an occupant type recognition module and an airbag ignition execution module. The occupant type recognition module includes a data acquisition module, an image processing module, and a classifier module;
[0055] like Figure 2 As shown, the data acquisition module includes a pressure sensor, a seat belt buckle sensor, an ultrasonic sensor, an image sensor and an acceleration sensor. There are two pressure sensors, which are respectively placed on the car seat and the car backrest to collect the body pressure of the occupant. The seat belt buckle sensor is placed at the seat belt buckle to collect the seat belt buckle status. The ultrasonic sensor is placed inside the roof just above the occupant to collect the distance between the occupant's head and the roof. The image sensor is a three-channel vehicle-mounted camera, which is placed above the windshield to collect the upper body image information of the passenger in the front passenger seat. The acceleration sensor is placed on the B-pillar of the car to collect the acceleration of the car.
[0056] The image processing module provides known samples and unknown samples for the learning and training of the classifier module by intercepting the region of interest, graying, filtering and smoothing, edge detection and feature extraction. The region of interest is intercepted: when the occupant on the seat cushion is not an infant in the applicable period of the safety seat, the image of the occupant at the corresponding frame number at time t is obtained through the on-board camera, and the image is cropped to narrow the recognition range to a smaller range where the occupant is located to improve the effectiveness of the information. The range should be adaptively adjusted according to the distance between the camera and the occupant;
[0057] Grayscale: Generally, vehicle-mounted cameras have three-channel images. The cropped images are grayscaled through the vehicle-mounted computer, and the three-channel images (red, green, and blue) are changed to two-channel images (black and white), which improves the recognition efficiency of the classifier and meets the real-time requirements.
[0058] Filtering and smoothing: Since the images collected by the vehicle-mounted camera are often affected by external factors, resulting in image distortion, the grayscale images are subjected to Gaussian filtering to remove noise from the image and improve the recognition success rate.
[0059] Edge detection: A gradient-based edge detection method is used to distinguish edge points from non-edge points using the characteristics of grayscale mutations at the edge of the image, and the Canny operator is used to separate the occupants from the image background.
[0060] Feature extraction: Legendre moments are used to describe the edge features of the occupants, thereby realizing the extraction of occupant feature space;
[0061] The classifier module is used to classify the gender and age of the occupants. The classifier adopts the support vector machine method. The SVM method is based on the VC dimension theory of statistical learning theory and the principle of structural risk minimization. It shows many unique advantages in solving small sample, nonlinear and high-dimensional pattern recognition.
[0062] The support vector machine includes two stages: learning and recognition. In the learning stage, the first step is to collect sample images of passengers, and then train the passenger model with known samples. The recognition stage is to detect unknown samples with the prediction model obtained after training, so as to classify the gender and age of passengers.
[0063] The airbag ignition execution module is used to ignite the airbag.
[0064] like Figure 4 As shown, a smart airbag ignition method based on occupant classification recognition includes the following steps:
[0065] S1. Determine the occupant type through the data acquisition module, the image processing module, and the classifier module, and calculate the ignition threshold A;
[0066] S2. Obtain the acceleration data of the vehicle through the acceleration sensor, where t1 is the collision start time, t2 is the best ignition time obtained after optimizing the image measurement results, and t3 = t2 + 30ms is the time when the airbag is fully filled with gas;
[0067] S3, detect and calculate the integral result S according to the time window width;
[0068] S4. When the integral result S is greater than or equal to the ignition threshold A, the airbag is ignited and the actual ignition time is recorded;
[0069] When the integration result S is less than the ignition threshold A, the airbag does not ignite, and the process returns to step S2.
[0070] like Figure 1 As shown, step S1 specifically includes the following steps:
[0071] S1.1:
[0072] The system starts to operate when the pressure sensor placed on the car seat is affected by a change in pressure;
[0073] S1.2:
[0074] During the operation of the system, if the pressure sensor at the seat back receives a pressure signal, the seat belt buckle sensor will detect whether the seat belt buckle is closed;
[0075] If the seat belt buckle is closed, the seat belt closing period obtained by the seat belt buckle sensor is used as the occupant's sitting posture correction time T;
[0076] When the passenger sits in a correct posture, the ultrasonic sensor placed on the top of the passenger works to measure the distance between the top of the passenger's head and the roof during the T period. The distances corresponding to different times are sent to the on-board computer, and the minimum distance from the top of the head to the roof is selected. This distance value is set as x1 in centimeters, and the time corresponding to x1 is set as t.
[0077] The distance from the roof to the seat cushion is set as x2, in centimeters;
[0078] The height l from the buttocks to the top of the head when sitting upright is x2-x1, in centimeters;
[0079] The average height of adult males / the average height from the buttocks to the top of the head of an adult male sitting upright is a, and in this embodiment a=1.923;
[0080] The predicted height of the occupant is L = 1.923*l, in centimeters;
[0081] S1.3:
[0082] The electronic signal obtained by the pressure sensor at the seat cushion is sent to the on-board computer. If the computer recognizes that the body pressure distribution image is a seamless regular pattern, it means that the seat cushion is in contact with the safety seat rather than the passenger, and the pending result 1 is obtained, that is, the passenger on the seat cushion is an infant in the applicable period of the safety seat;
[0083] S1.4:
[0084] When the occupant on the seat cushion is not an infant within the applicable period of the safety seat, the vehicle-mounted camera obtains a picture of the occupant at the corresponding frame number at time t, crops the picture, and narrows the recognition range to a smaller range where the occupant is located to improve the effectiveness of the information. The range should be adaptively adjusted according to the distance between the camera and the occupant;
[0085] Generally, the on-board camera has three-channel images. The cropped images are grayed out by the on-board computer, and the three-channel images (red, green, and blue) are changed to two-channel images (black and white), which improves the recognition efficiency of the classifier and meets the real-time requirements.
[0086] Since the images collected by the on-board camera are often affected by external factors, resulting in image distortion, the grayscale images are subjected to Gaussian filtering to remove noise in the image and improve the recognition success rate;
[0087] S1.5:
[0088] Use the trained classifier to detect the gender and age of the occupants on the processed image samples;
[0089] If the occupant is detected to be younger than 12 years old, the pending result 1 is changed to the real result 1, and the occupant is an infant within the applicable period of the safety seat;
[0090] If the age of the occupant is greater than 12 years old but less than 18 years old, the result is 2, and the occupant is a child;
[0091] If the age is greater than 18 years old and the deep neural network classifier determines that the occupant's gender is female, the result 3 is obtained, and the occupant is an adult female;
[0092] If the age is greater than 18 years old and the gender is male, the predicted height L of the occupant obtained in S3.2 is used for judgment;
[0093] If the occupant's predicted height L is greater than 188 cm, the result is 4, and the occupant is a 95th percentile male;
[0094] If the occupant's predicted height L is less than 188 cm, the result is 5, and the occupant is a 50th percentile male;
[0095] S1.6:
[0096] The acceleration data of the car is obtained through the acceleration sensor, where t1 is the start time of the collision, t2 is the best ignition time obtained after optimizing the image measurement results, and t3=t2+30ms is the time when the airbag is fully filled with gas.
[0097] In step S3, at time t2, the algorithm should be very sensitive to the intensity of the collision and be able to distinguish different collision intensities to determine whether to deploy the airbag, while also having good anti-interference capabilities. During the collision, the wider the time window, the greater the speed change, and the narrower the time window, the smaller the speed change. In order to evaluate and select a suitable time window width w, the enumeration method is used to select window widths of 20ms, 30ms, 40ms, 50ms, 60ms, and 70ms. Divide by the window width w to get A(t,w). A(t,w) represents the intensity of the collision within the time window. The larger the value, the stronger the collision. In order to analyze the sensitivity and anti-interference ability of the collision curve under different window widths, A(t,w) is normalized, that is, A(t,w) is divided by the maximum value A(t,w). max , and obtain the normalized index
[0098] Figure 3 It is the result of processing the collision test data with a collision speed of 50km / h according to the normalized index. t1 is the target ignition time, which can be obtained through a high-speed camera in a real car collision according to the 5in-30ms airbag ignition control criterion. t2 is the moment when the airbag is fully filled with gas under the condition of target ignition. The time windows of curves L1, L2, L3, L4, L5, and L6 are 20ms, 30ms, 40ms, 50ms, 60ms, and 70ms respectively. Curve L6 has a value exceeding that of moment t1 before moment t1, which means that when the window width is greater than 60ms, pre-ignition will occur, which is easy to cause mis-ignition. Curves L1, L2, L3, L4, and L5 do not exceed the value of moment t1 before moment t1, and they are monotonically rising, which means that target ignition can be achieved when the window height reaches the threshold. At time t1, the slope of curve L3 is the largest, which indicates that the algorithm is more sensitive to collision when the window width is 40ms, and the appropriate window width is finally selected as 40ms.
[0099] Taking the 95th percentile adult male as an example, the ignition threshold A corresponding to the 95th percentile adult male under a 40ms window width is obtained according to the above steps. If the occupant detection result is the 95th percentile adult male, the acceleration data is integrated in real time under a 40ms window width:
[0100] When the integral result S is greater than or equal to the ignition threshold A, the airbag ignites and the actual ignition time is recorded;
[0101] When the integration result S is less than the ignition threshold A, the airbag does not ignite and continues to integrate the acceleration data in real time with a window width of 40ms.
[0102] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. An intelligent airbag ignition system based on occupant classification recognition, characterized in that It includes an occupant type recognition module and an airbag ignition algorithm module. The occupant type recognition module includes a data acquisition module, an image processing module, and a classifier module; The data acquisition module includes a pressure sensor, a seat belt buckle sensor, an ultrasonic sensor, an image sensor and an acceleration sensor. There are two pressure sensors, which are respectively placed on the car seat and the car backrest to collect the body pressure of the occupant. The seat belt buckle sensor is placed at the seat belt buckle to collect the seat belt buckle status. The ultrasonic sensor is placed inside the roof just above the occupant to collect the distance between the occupant's head and the roof. The image sensor is a three-channel vehicle-mounted camera, which is placed above the windshield to collect the upper body image information of the passenger in the front passenger seat. The acceleration sensor is placed on the B-pillar of the car to collect the acceleration of the car. The image processing module provides known samples and unknown samples for the learning and training of the classifier module by intercepting the region of interest, graying, filtering and smoothing, edge detection and feature extraction; The classifier module classifier is used to classify the occupants in terms of gender and age; The airbag ignition algorithm module is used to design different ignition thresholds according to different types of occupant collision optimization parameters.
2. An intelligent airbag ignition method based on occupant classification recognition, characterized in that: The following steps are involved: S1. Determine the occupant type through the data acquisition module, the image processing module, and the classifier module, and calculate the ignition threshold A; S2. Obtain the acceleration data of the vehicle through the acceleration sensor, where t1 is the collision start time, t2 is the best ignition time obtained after optimizing the image measurement results, and t3 = t2 + 30ms is the time when the airbag is fully filled with gas; S3, detect and calculate the integral result S according to the time window width; S4. When the integral result S is greater than or equal to the ignition threshold A, the airbag is ignited and the actual ignition time is recorded; When the integration result S is less than the ignition threshold A, the airbag does not ignite, and the process returns to step S2.
3. The intelligent airbag ignition method based on occupant classification recognition according to claim 2 is characterized in that: The following steps are involved: S1.1: The system starts to operate when the pressure sensor placed on the car seat is affected by a change in pressure; S1.2: During the operation of the system, if the pressure sensor at the seat back receives a pressure signal, the seat belt buckle sensor will detect whether the seat belt buckle is closed; If the seat belt buckle is closed, the seat belt closing period obtained by the seat belt buckle sensor is used as the occupant's sitting posture correction time T; When the passenger sits in a correct posture, the ultrasonic sensor placed on the top of the passenger works to measure the distance between the top of the passenger's head and the roof during the T period. The distances corresponding to different times are sent to the on-board computer, and the minimum distance from the top of the head to the roof is selected. This distance value is set as x1 in centimeters, and the time corresponding to x1 is set as t. The distance from the roof to the seat cushion is set as x2, in centimeters; The height l from the buttocks to the top of the head when sitting upright is x2-x1, in centimeters; The average height of adult males / the average height from hips to head of adult males sitting upright is a; The predicted height of the occupant is L = a*l, in centimeters; S1.3: The electronic signal obtained by the pressure sensor at the seat cushion is sent to the on-board computer. If the computer recognizes that the body pressure distribution image is a seamless regular pattern, it means that the seat cushion is in contact with the safety seat rather than the passenger, and the pending result 1 is obtained, that is, the passenger on the seat cushion is an infant in the applicable period of the safety seat; S1.4: When the occupant on the seat cushion is not an infant within the applicable period of the safety seat, a picture of the occupant at a corresponding frame number at time t is obtained through the vehicle-mounted camera, and the picture is cropped; S1.5: Use the trained classifier to detect the gender and age of the occupants on the processed image samples; If the occupant is detected to be younger than 12 years old, the pending result 1 is changed to the real result 1, and the occupant is an infant within the applicable period of the safety seat; If the age of the occupant is greater than 12 years old but less than 18 years old, the result is 2, and the occupant is a child; If the age is greater than 18 years old and the deep neural network classifier determines that the occupant's gender is female, the result 3 is obtained, and the occupant is an adult female; If the age is greater than 18 years old and the gender is male, the predicted height L of the occupant obtained in S3.2 is used for judgment; If the occupant's predicted height L is greater than 188 cm, the result is 4, and the occupant is a 95th percentile male; If the occupant's predicted height L is less than 188 cm, the result is 5, and the occupant is a 50th percentile male; S1.6: The acceleration data of the car is obtained through the acceleration sensor, where t1 is the start time of the collision, t2 is the best ignition time obtained after optimizing the image measurement results, and t3=t2+30ms is the time when the airbag is fully filled with gas.
4. The intelligent airbag ignition method based on occupant classification recognition according to claim 2 is characterized in that: In step S3, the window widths are selected as 20ms, 30ms, 40ms, 50ms, 60ms, and 70ms respectively by enumeration method, and the formula Divide by the time window width w to get A(t,w), normalize A(t,w), and divide A(t,w) by the maximum value A(t,w) max , and obtain the normalized index
Citation Information
Cited By
Vehicle rolling protection method and device, medium and electronic equipment
CN119773669A
