A car active energy absorption device based on pedestrian classification recognition and a control method thereof

Abstract

The application discloses an automobile active energy-absorbing device based on pedestrian classification and recognition and a control method thereof, wherein the automobile active energy-absorbing device is an adult / child airbag structure, the method comprises the following steps: acquiring a real-time road image; inputting the real-time road image into a pedestrian classification model based on a YOLOv5 network for classification detection to obtain a pedestrian classification result, that is, identifying whether a pedestrian exists and whether the pedestrian is an adult or a child; detecting the distance between the pedestrian and the automobile and acquiring the current vehicle state, and based on the distance and the current vehicle state, it is predicted that a collision will occur, and according to the pedestrian classification result, the automobile active energy-absorbing device pops out a corresponding adult airbag or child airbag. The technical scheme of the application can avoid over-inflation, causing secondary injury to children, and insufficient inflation, failing to effectively protect the head of an adult, and at the same time, the whole airbag point explosion and opening process is realized before the real contact between the pedestrian and the automobile, thereby gaining valuable time for pedestrian protection.

Description

Technical Field

[0001] This invention belongs to the field of automotive energy absorption device technology, specifically relating to an active automotive energy absorption device and its control method based on pedestrian classification and recognition. Background Technology

[0002] The advent of the automobile has dramatically changed the way people travel, but it has also brought many problems that cannot be ignored. Pedestrians, as a vulnerable group in the traffic environment, have drawn widespread attention to their safety. To prevent pedestrian-vehicle collisions, combining pedestrian detection technology with active energy absorption devices in automobiles is a key technology for reducing pedestrian hazards. This is significant for improving the active safety protection capabilities of automobiles and also has practical implications for reducing additional expenses such as vehicle maintenance and repairs.

[0003] Currently, E. Bergenheim and others from Volvo Cars have filed a patent for "a pedestrian protection airbag for a vehicle". The airbag deploys along the windshield and / or A-pillar. When the vehicle's sensor system detects a potential collision risk to a pedestrian or cyclist, the airbag deploys and inflates to cover the windshield and / or A-pillar, so that the pedestrian / cyclist will collide with the soft airbag.

[0004] Wu Taide et al. filed an invention patent for "A Pedestrian Protection System," disclosing a vehicle and pedestrian protection system that deploys external airbags to block and / or cushion contact between the airbag-equipped vehicle and pedestrians, animals, other vehicles, and / or stationary objects. During a collision, the pedestrian and vehicle protection system uses a controller with one or more sensors to determine when the airbags should deploy, utilizes sensor feedback to detect hazardous situations, and assesses the severity of the hazardous situation to activate one or more airbags.

[0005] However, of the two technical solutions mentioned above, the first solution designs a pedestrian protection airbag device but does not specifically implement pedestrian classification and targeted processing. For pedestrians such as children, if the deployed pedestrian protection airbag inflates too much, it may cause secondary injury to the child; for pedestrians such as adults, if the deployed pedestrian protection airbag inflates too little, it fails to achieve the protective performance of the airbag. The second solution uses one or more sensors to detect pedestrians or other collision objects. However, these sensors can only receive collision information after a collision has occurred. They then analyze and process the received information, and finally the vehicle control module decides whether to deploy the airbag based on the analysis results. This results in a higher requirement for the airbag deployment time, which has a lag and still poses a significant risk to pedestrians. Summary of the Invention

[0006] This invention addresses the problems of existing technologies by providing an active energy-absorbing vehicle device and its control method based on pedestrian classification and recognition. The method categorizes pedestrians into adults and children, and uses these categories to create two different types of inflatable airbags. This avoids over-inflation, which could cause secondary injury to children, and under-inflation, which could fail to effectively protect the head of adults. Furthermore, this invention incorporates a YOLOv5 network to construct a pedestrian classification model for classification and detection. This allows for early warning assessment of dangerous situations before a pedestrian-vehicle collision, buying valuable time for pedestrian protection and ensuring the entire airbag deployment process occurs before actual contact between the pedestrian and vehicle.

[0007] Therefore, the technical solution of the present invention is as follows:

[0008] On one hand, the present invention provides a control method for an active energy-absorbing device for automobiles based on pedestrian classification and recognition. The active energy-absorbing device is located at the front of the automobile and is an adult / child airbag structure. The method includes:

[0009] Real-time images of the road are obtained by using a vehicle-mounted camera;

[0010] The real-time road image is input into a pedestrian classification model built on a YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children.

[0011] The distance between pedestrians and cars is detected by using vehicle-mounted sensing devices and the current vehicle status is obtained. Based on the distance between pedestrians and cars and the current vehicle status, it is predicted whether a collision between pedestrians and cars will occur.

[0012] If a collision is predicted, the vehicle's active energy absorption device deploys the corresponding adult or child airbag based on the pedestrian classification results; otherwise, monitoring continues.

[0013] Further optionally, the YOLOv5 network includes an input terminal, a backbone network, a neck network, and an output terminal connected in sequence.

[0014] Specifically, the SPP module in the traditional backbone network is replaced with Spatial Pyramid Pooling (SPPCSPC) to enhance network robustness and improve the anti-interference level of child target recognition. The SPPCSPC is divided into two transmission paths. The first transmission path has three convolutional layers (CBS), a max pooling layer, a feature fusion module, and two convolutional layers (CBS) in sequence. The second transmission path has a convolutional layer (CBS). The output of the last convolutional layer (CBS) in the first transmission path is fused with the output of the last convolutional layer (CBS) in the second transmission path, and then input into the convolutional layer (CBS).

[0015] The technical solution of this invention takes into account that pedestrian detection is affected by factors such as lighting conditions, clothing, and occlusion, which can reduce the effectiveness of the network and lead to false detections. Furthermore, adults and children have different scales within an image; small pedestrian targets generally have lower resolution, smaller pixel area, and are more susceptible to noise interference, thus preventing the network from accurately locating pedestrians. To make the network less susceptible to interference, the SPPC module is introduced to replace the SPP module, enhancing the network's robustness.

[0016] Further optionally, the YOLOv5 network includes an input terminal, a backbone network, a neck network, and an output terminal connected in sequence.

[0017] In this system, the first C3 module on the backbone network has another output path, and the second C3 module on the backbone network has another output path. The two paths are merged, and then pass through the Focus layer, the Convolutional Layer (CBS), and the Feature Fusion module in sequence before being connected to the traditional transmission path of the second C3 module to the neck network.

[0018] The technical solution of this invention takes into account the following: In road traffic environments, pedestrians are in a dynamic state, resulting in varying sizes for adults and smaller sizes for children compared to adults. This causes small target objects to gradually lose their pedestrian feature information in the image as the network depth increases, leading to a gradual decrease in recognizability. Only a rough outline of the pedestrian becomes visible, and it becomes increasingly blurry, unable to be distinguished from the background. Eventually, the feature map extracted by the SPPC module can no longer identify the pedestrian's outline. To enhance the network's ability to acquire information about small target pedestrian objects, a new path is added.

[0019] As can be seen from the above, in order to improve the recognition ability of adult and child targets, the technical solution of the present invention improves the network based on the features of real-time road images.

[0020] Further optionally, the current vehicle state includes at least the current vehicle speed, and the process of predicting whether a collision between a pedestrian and a car will occur based on the distance and the current vehicle state is as follows:

[0021] If the pedestrian is an adult and the current vehicle speed exceeds a preset vehicle speed threshold V1 and the distance is less than a preset distance threshold L1, it is determined that the adult and the car will collide.

[0022] If the pedestrian is a child and the current vehicle speed exceeds a preset vehicle speed threshold V2 and the distance is less than a preset distance threshold L2, it is determined that the child and the car will collide.

[0023] The preset vehicle speed threshold V1 and the preset vehicle speed threshold V2 are in the range of 5.6 m / s to 11.1 m / s.

[0024] The preset distance threshold L1 and the preset distance threshold L2 are in the range of 6.7m to 13.3m.

[0025] Further optionally, the relationship between the gas volume deployed by the adult or child airbag and the head shape acceleration and head shape mass is as follows:

[0026]

[0027] In the formula, P is the gas pressure; V is the gas volume; n is the gas quantity; R is the gas constant; T is the gas temperature; m is the head mass; S is the area of ​​the head in the direction of the velocity vector; and a is the head acceleration.

[0028] The gas volume of the corresponding adult-type airbag and child-type airbag is determined based on the aforementioned relationship and the approximately equivalent head mass and head acceleration of adults and children.

[0029] Secondly, the present invention also provides a control system based on the above method, comprising:

[0030] The image acquisition module is used to acquire real-time images of the road.

[0031] The pedestrian classification module is used to input the real-time road image into a pedestrian classification model built based on the YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children.

[0032] The collision detection module is used to obtain the distance between the pedestrian and the car and the current vehicle status, and to predict whether a collision between the pedestrian and the car will occur based on the distance between the pedestrian and the car and the current vehicle status.

[0033] The control module is used to deploy the corresponding adult or child airbags from the vehicle's active energy absorption device if a collision is predicted, based on the pedestrian classification results; otherwise, monitoring continues.

[0034] In three aspects, the present invention provides an active energy absorption device for automobiles based on the above method. The active energy absorption device for automobiles includes a child-type airbag and an adult-type airbag, or the active energy absorption device for automobiles includes an airbag and an airbag inflation device. The airbag inflation device controls the gas volume in the airbag according to the child or adult type to obtain a child-type airbag or an adult-type airbag.

[0035] In four aspects, the vehicle system based on the above method provided by the technical solution of the present invention includes at least: a vehicle controller, a vehicle active energy absorption device, a vehicle camera, and a vehicle sensing device;

[0036] Among them, the vehicle active energy absorption device, vehicle camera, and vehicle sensing device are all connected to the vehicle controller, and the vehicle controller calls or loads the pedestrian classification model built based on the YOLOv5 network.

[0037] The vehicle-mounted camera transmits real-time images of the road it captures to the vehicle controller.

[0038] The vehicle controller uses the pedestrian classification model to classify and detect the real-time road image to obtain the pedestrian classification result.

[0039] The vehicle-mounted sensing device detects the distance between pedestrians and the vehicle and obtains the current vehicle status, and transmits it to the vehicle controller;

[0040] The vehicle controller predicts whether a collision will occur based on the distance between the pedestrian and the vehicle and the current vehicle status. If a collision is predicted, the vehicle's active energy absorption device deploys the corresponding adult or child airbag according to the pedestrian classification result; otherwise, monitoring continues.

[0041] In five aspects, the present invention provides a computer-readable storage medium storing a computer program, which is invoked by a processor to implement the following:

[0042] Acquire real-time images of the road;

[0043] The real-time road image is input into a pedestrian classification model built on a YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children.

[0044] The distance between the pedestrian and the car, as well as the current vehicle status, are obtained, and based on the distance and the current vehicle status, it is predicted whether a collision between the pedestrian and the car will occur.

[0045] If a collision is predicted, the vehicle's active energy absorption device deploys the corresponding adult or child airbag based on the pedestrian classification results; otherwise, monitoring continues.

[0046] Beneficial effects

[0047] Compared with existing methods, the advantages of the present invention are:

[0048] 1. The control method provided by this invention, on the one hand, classifies pedestrians into adults and children, and deploys adult-type airbags and child-type airbags respectively for adults and children, avoiding over-inflation which could cause secondary injury to children, and under-inflation which could fail to effectively protect the adult's head. On the other hand, this invention introduces a YOLOv5 network to construct a pedestrian classification model for classification and detection, enabling early warning judgment of dangerous situations before a pedestrian-vehicle collision, buying valuable time for pedestrian protection, and realizing the entire airbag deployment process before the actual contact between the pedestrian and the vehicle, thus more effectively reducing the collision damage.

[0049] 2. This invention optimizes the traditional YOLOv5 network. To enhance the network's ability to acquire feature information of small targets, this invention utilizes feature information from shallow layers to enhance the fusion of deep and shallow features in the YOLOv5 network. Therefore, the feature layers output by the first C3 module are fused with those output by the second C3 module. To fuse feature layers of different scales, this invention employs a focus layer, aiming to achieve a 2x downsampling without losing feature information. This invention introduces Spatial Pyramid Pooling (SPPC) to replace the SPP module in the original YOLOv5 network. In SPPC, SPP contains multiple different MaxPools, giving the network multiple receptive fields to distinguish target objects of different sizes. In CSP, the feature layer is first divided into two parts: one part undergoes conventional processing such as convolution, normalization, and activation functions, while the other part undergoes SPP structure processing. Finally, these two parts are superimposed using Concat, which helps reduce the computational load by half, improving both speed and accuracy. Since pedestrians are affected by various factors such as color, shape, and environment, this invention also introduces a CA attention mechanism to extract features that are more beneficial to network training. The purpose of this mechanism is to tell the network which content and which locations need to be paid more attention to.

[0050] 3. The technical solution of the present invention approximately controls the peak acceleration of the head impact airbag by controlling the airbag deployment volume or the amount of gas in the airbag, thereby reducing the damage to the head. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the airbag deployment of an active energy absorption system in a car.

[0052] Figure 2 This is a diagram of the YOLOv5 network architecture;

[0053] Figure 3 This is an architecture diagram of some modules of the YOLOv5 network;

[0054] Figure 4 This is the mass flow curve for an adult airbag;

[0055] Figure 5 This is the mass flow curve of a child airbag;

[0056] Figure 6 These are the results of adult head impact tests. Figures a and b are schematic diagrams of an adult head impacting the front of a car without an airbag, and corresponding adult head impact HIC diagrams, respectively. Figures c and d are schematic diagrams of an adult head impacting the front of a car equipped with an adult head airbag, and corresponding adult head impact HIC diagrams, respectively. Figures e and f are schematic diagrams of an adult head impacting the front of a car equipped with a child head airbag, and corresponding adult head impact HIC diagrams, respectively.

[0057] Figure 7 These are the results of the child head impact test. Figures a and b are schematic diagrams of a child impacting the front of a car without an airbag, and the corresponding child head shape HIC diagrams, respectively. Figures c and d are schematic diagrams of a child impacting the front of a car equipped with an adult airbag, and the corresponding child head shape HIC diagrams, respectively. Figures e and f are schematic diagrams of a child impacting the front of a car equipped with a child airbag, and the corresponding child head shape HIC diagrams, respectively. Detailed Implementation

[0058] This invention provides an active energy-absorbing vehicle device and its control method based on pedestrian classification and recognition. For the first time, it proposes classifying pedestrians and deploying corresponding airbags for adults and children, avoiding over-inflation which could cause secondary injury to children, and under-inflation which would fail to effectively protect adults' heads. Furthermore, it introduces a YOLOv5 network to construct a pedestrian classification model for classification and detection, enabling early warning judgments of dangerous situations before a pedestrian-vehicle collision, thus realizing the entire airbag deployment process before actual contact between the pedestrian and vehicle. The invention will be further described below with reference to embodiments.

[0059] like Figure 1 As shown, this invention includes an active energy-absorbing device on the front of a car, specifically an adult / child airbag structure. After deployment, the airbag covers dangerous areas such as the rear edge of the hood, the front edge of the windshield, and the A-pillar. A pedestrian's head will collide with the soft airbag, thereby reducing the risk of head injury. In some embodiments, the active energy-absorbing device includes a child-type airbag and an adult-type airbag. That is, it has two airbags, with the corresponding adult airbag deployed selectively. In some embodiments, the active energy-absorbing device includes an airbag and an airbag inflation device. The airbag inflation device controls the gas volume inside the airbag according to whether it is a child or an adult airbag, thus creating either a child-type or adult-type airbag.

[0060] Based on an active energy absorption device for automobiles, the present invention provides a control method for an active energy absorption device for automobiles based on pedestrian classification and recognition, comprising the following steps:

[0061] Real-time images of the road are obtained by using a vehicle-mounted camera;

[0062] The real-time road image is input into a pedestrian classification model built on a YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children.

[0063] The distance between pedestrians and cars is detected by using vehicle-mounted sensing devices and the current vehicle status is obtained. Based on the distance and the current vehicle status, it is predicted whether a collision between pedestrians and cars will occur.

[0064] If a collision is predicted, the vehicle's active energy absorption device deploys the corresponding adult or child airbag based on the pedestrian classification results; otherwise, monitoring continues.

[0065] Among them, such as Figure 2 As shown, the YOLOv5 network architecture of the pedestrian classification model provided by the technical solution of this invention is as follows:

[0066] The improved YOLOv5 network comprises an input terminal, a backbone network, a neck network, and an output terminal (head). The input terminal preprocesses the real-time road images. In this embodiment, the Mosaic data augmentation method is employed, which randomly selects four images from the dataset during training, performs rotation and scaling operations, and then randomly stitches them together to form a new training set, thereby enhancing the network's robustness and generalization ability. In other feasible embodiments, the preprocessing techniques are not limited to the examples mentioned above; any preprocessing method that enhances image features can be considered within the scope of this invention.

[0067] The backbone network adopts the traditional architecture but has been optimized based on the features of real-time road images and the characteristics of adults and children, improving the recognition ability for adults and children. Specifically, it includes a Focus layer, a Convolutional Baseline Stream (CBS) layer (Conv2D_BN_SiLU), a C3 module, a CA attention module, and a Spatial Pyramid Pooling (SPPCSPC) module. In this embodiment, the Focus layer, two CBS layers, the first C3 module, a CBS layer, a second C3 module, a CBS layer, a third C3 module, a CBS layer, a CA attention module, the SPPCSPC module, and the fourth C3 module are connected sequentially. This invention replaces the SPP module of the traditional backbone network with the SPPCSPC module.

[0068] The Focus layer performs a slicing operation, downsampling the input feature layers. This means taking the value of every single pixel in an image, resulting in four independent feature layers. This expands the original image's RGB 3 channels by a factor of four, while simultaneously reducing the image's height and width by half. Its advantage lies in minimizing information loss and reducing computational load while performing downsampling. The C3 module maps the image features into two parts: one part undergoes residual convolution, and the other part is processed by a CBS convolutional layer. The two parts are then concatenated, separating shallow features and mitigating the vanishing gradient problem.

[0069] Since pedestrians are influenced by various factors such as color, shape, and environment, this invention introduces a CA attention mechanism to extract features that are more beneficial for network training. The purpose of this mechanism is to tell the network which content and locations require more attention. The CA attention mechanism first performs global average pooling on both the height and width, then uses convolution and concatenation to reduce the dimensionality of the input image, followed by dimensionality up-multiplication to obtain attention weights in both directions. Finally, the obtained weights are multiplied and weighted with the input feature map.

[0070] Traditional YOLOv5 uses the SPP module, which employs three different max-pooling layers to enhance the network's receptive field. To enable the network to extract more reliable feature information and improve detection accuracy, in CSP, the feature layer is first divided into two parts. One part undergoes conventional processing such as convolution, normalization, and activation functions, while the other part undergoes SPP structure processing. Finally, these two parts are concatenated using a concat operation, which helps reduce the computational load by half, thus improving both speed and accuracy. This invention introduces Spatial Pyramid Pooling (SPPC) to replace the SPP module in the original YOLOv5 network. SPPC consists of two transmission paths. The first transmission path sequentially includes three convolutional layers (CBS), a max-pooling layer, a feature fusion module, and two more convolutional layers (CBS). The second transmission path includes a convolutional layer (CBS). The output of the last convolutional layer (CBS) in the first transmission path is fused with the output of the last convolutional layer (CBS) in the second transmission path, and then input into the second convolutional layer (CBS). Among them, the convolutional layer (CBS) is a basic component of the network, which consists of convolution, normalization, and activation functions.

[0071] The neck network, also known as an enhanced feature extraction network, employs a feature pyramid and path aggregation network structure. The feature pyramid structure fuses high-level and shallow semantic information through upsampling, thereby enhancing the network's feature learning ability. The path aggregation network structure adds a bottom-up feature fusion path to the feature pyramid structure, allowing shallow information to be better transmitted to higher levels, further improving the network's detection performance. This invention optimizes the traditional neck network by optimizing the transmission path of the second C3 module to the traditional neck network. Specifically, another output path is derived from the first C3 module, and another output path is derived from the second C3 module. These two paths are fused, and then sequentially passed through the Focus layer, the Convolutional Baseline (CBS) layer, and the feature fusion module before being connected to the traditional neck network. It should be noted that as the network depth increases, after multiple convolutional operations, the size of the output feature map continuously shrinks, ultimately causing the loss of small target feature information contained in the feature map during transmission. Therefore, to enhance the network's ability to acquire feature information of small targets, this invention utilizes the feature information of shallow networks to enhance the fusion of deep and shallow features in the YOLOv5 network. Thus, the feature layers output by the first C3 module are fused with those output by the second C3 module. To fuse feature layers of different scales, this invention employs a focus layer, which aims to achieve a 2x downsampling without losing feature information. The focus layer transforms the 160*160 feature layer into an 80*80 feature layer. Then, standard convolution is used to match the number of channels in the feature layer with the output channels of the second C3 module. Finally, a concat operation is used to complete the feature fusion.

[0072] The output is the classification result. In this embodiment, the GIOU function is used as the loss function for bounding boxes, and non-maximum suppression is used to filter multiple target boxes.

[0073] In this embodiment of the invention, a modified YOLOv5 network is preferably used to construct a pedestrian classification model. The input data of this pedestrian classification model is real-time road images, and the output results are pedestrian recognition results and pedestrian categories (adults and children). In order to use the pedestrian classification network for pedestrian recognition, the network needs to be trained in advance. The training process will be illustrated below with an example:

[0074] 1. Prepare the sample dataset. Using the Labelimg software, the existing target dataset (composed of road images) was selected, modified, and re-labeled (pedestrian annotations and pedestrian category annotations). The re-categorized classes are adults and children. The dataset contains 14,000 images, including 21,070 adult-annotated samples and 6,600 child-annotated samples, encompassing diverse pedestrian image samples under different geographical environments, weather conditions, light and shadow variations, and gender differences.

[0075] 2. Next, train the YOLOv5 model on a PC. Input the prepared dataset, iteratively update the network parameters through propagation, and calculate the loss function. Stop training when the number of training iterations is reached or the set threshold is met.

[0076] 3. After training, the training results are evaluated based on the evaluation metrics. During training, the IOU value of YOLOv5 is set to 0.5. If the threshold exceeds 0.5, it means the current target was correctly identified, and the predicted bounding box will be retained; if the threshold does not exceed 0.5, it means the current target was incorrectly identified, and the predicted bounding box will not be retained. In YOLOv5, the higher the precision, the fewer false positives. The higher the recall, the fewer false negatives.

[0077] It should be noted that, as Figure 3 As shown, the training process of the YOLOv5 network is a process implemented using existing technology, and the loss function involved can also refer to existing technology; this invention does not impose specific limitations on it. The trained pedestrian classification model is then loaded into the vehicle controller.

[0078] After the acquired real-time road images are input into the pedestrian classification model, the pedestrian classification results are obtained. Simultaneously or subsequently, the distance between pedestrians and vehicles is detected and the current vehicle status is obtained using vehicle-mounted sensing devices. Based on the distance and the current vehicle status, it is then predicted whether a collision between pedestrians and vehicles will occur.

[0079] In this embodiment, if the current pedestrian is an adult, the vehicle speed exceeds 8.52 m / s, and the distance is less than 10 m, the vehicle control module will send a command to the pedestrian protection module to deploy the adult airbag; if the current pedestrian is a child, the vehicle speed exceeds 8.52 m / s, and the distance is less than 10 m, the vehicle control module will send a command to the pedestrian protection module to deploy the child airbag; if the distance between the adult / child and the vehicle is greater than 10 m and the vehicle speed is less than 8.52 m / s, the pedestrian airbag will not deploy.

[0080] Regarding the design of adult and child airbags, this invention has discovered that the relationship between the deployed gas volume of an adult or child airbag and the head shape acceleration and head shape mass is as follows:

[0081]

[0082] In the formula, P is the gas pressure; V is the gas volume; n is the gas quantity; R is the gas constant; T is the gas temperature; m is the head mass; S is the area of ​​the head in the direction of the velocity vector; and a is the head acceleration.

[0083] Specifically, based on the aforementioned relationship and the approximately equivalent head mass and head acceleration for adults and children, the gas volume for deploying the corresponding adult and child airbags is determined. This allows for the design / selection of the appropriate airbags to effectively protect the heads of both adults and children. In other words, the equivalent head mass can be determined based on the average level of children / adults, and the approximately equivalent head acceleration is set according to application requirements and to minimize harm.

[0084] This invention employs the uniform pressure method to simulate an active energy absorption device in a car. This method uses a volume control approach, describing the gas generated by the gas generator through mass flow rate and temperature. In the uniform pressure model, it is assumed that the gas inflated into the airbag is ideal and that the airbag is thermally insulated from the outside environment, thus:

[0085]

[0086] In the formula, P2 is the pressure of the airbag; V2 is the volume of the airbag; m2 is the mass of the gas inside the airbag; T2 is the temperature of the gas inside the airbag; R is the ideal gas constant; ρ is the gas density; e is the gas energy; and k is the heat capacity ratio constant.

[0087] The rate of change of the gas mass inside the control volume is determined by the gas mass flowing through the boundary at time t, that is:

[0088]

[0089] In the formula, m min The mass flow rate of gas flowing into the airbag is m, which is the mass flow rate of gas from the gas generator into the airbag. m12 m mout The mass flow rate of gas exiting the airbag is m, which is the mass flow rate m exiting through the exhaust port. m23 and the leaked mass flow rate m' m23 sum.

[0090] Based on the above formula, the volume, pressure, and other state parameters of the airbag at each moment during deployment can be determined, such as... Figure 4 as well as Figure 5This allows for the approximate control of the peak acceleration of a head-impact airbag by controlling the airbag's deployment volume or gas quantity, thereby reducing head injury. Similarly, after determining the airbag's volume, pressure, and other state parameters at each moment during deployment, the airbag volume can be controlled to further reduce injury based on safety requirements for other state parameters.

[0091] Impact test analysis: Based on simulation results Figure 6 It can be seen that in the adult model, such as Figure 6 Figures a and b show that when an adult head-shaped impactor strikes the frontal structure of a car without an active energy absorption device, the peak acceleration is approximately 280.72g, and the head HIC value reaches as high as 1682.67, causing severe damage to the adult's head. Figure 6 Figures c and d show that when an adult head-shaped impactor strikes a car equipped with an adult-shaped active energy absorption device, the peak acceleration is 115.87g, and the head HIC value is 820.895, representing a decrease of 51.2% and 58.7% in peak acceleration and HIC value, respectively. Figure 6 Figures e and f show that when an adult head-shaped impactor collides with a child-sized active energy-absorbing device in a car, the peak acceleration is 238.54g, and the head HIC value is 1181.06, representing a reduction of 15.0% and 29.8% in peak acceleration and HIC value, respectively. Compared to child-sized active energy-absorbing devices, adult-sized active energy-absorbing devices significantly reduce head injuries in adults, providing better protection for their heads.

[0092] Based on simulation results Figure 7 It can be seen that: Figure 7 Figures a and b show that, in a child model, when a child's head impacts the frontal structure of a car without an active energy absorption device, the peak acceleration is approximately 196.38g, and the head HIC value is 761.01. Figure 7 Figures c and d show that when a child's head-shaped impactor collides with an active energy-absorbing device in an adult-sized car, the peak acceleration is 167.22g. However, because the active energy-absorbing device in the adult-sized car fails to effectively absorb the impact energy of the child's head-shaped impactor, the acceleration decreases slowly after reaching its peak. Even after the acceleration decreases, it is still higher than the acceleration when impacting without an active energy-absorbing device, ultimately causing the head's HIC value to rise to 1320.07, resulting in more severe head injury for the child. Figure 7Figures e and f show that when a child's head impactor strikes a child-sized active energy-absorbing device in a car, the peak acceleration is 91.93g, and the head HIC value is 522.76, representing a reduction of 53.2% and 31.3% in peak acceleration and HIC value, respectively. Compared to active energy-absorbing devices for adults, active energy-absorbing devices for children reduce head injuries and provide better protection for children's heads.

[0093] In some embodiments, the present invention provides a system based on the above control method, which includes at least: an image acquisition module, a pedestrian classification module, a collision judgment module, and a control module.

[0094] The system includes: an image acquisition module for acquiring real-time road images; a pedestrian classification module for inputting the real-time road images into a pedestrian classification model built on a YOLOv5 network for classification and detection to obtain pedestrian classification results, which are used to identify whether pedestrians exist in the real-time road images and whether they are adults or children; a collision judgment module for acquiring the distance between the pedestrian and the vehicle and the current vehicle status, and predicting whether a collision between the pedestrian and the vehicle will occur based on the distance and the current vehicle status; and a control module for deploying the corresponding adult or child airbags from the vehicle's active energy absorption device if a collision is predicted, based on the pedestrian classification results; otherwise, monitoring continues.

[0095] It should be noted that the above embodiments describe the technical solution of the present invention from the perspective of functional modules. It should be understood that the implementation process of each module can refer to the content description of the foregoing method. The above division of functional modules is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. At the same time, the above-mentioned integrated units can be implemented in hardware or in the form of software functional units.

[0096] In some embodiments, the present invention provides a system based on the above-described cabinet system method, which includes at least: an automobile controller, an automobile active energy absorption device, an in-vehicle camera, and an in-vehicle sensing device.

[0097] Among them, the vehicle active energy absorption device, vehicle camera, and vehicle sensing device are all connected to the vehicle controller, and the vehicle controller calls or loads the pedestrian classification model built based on the YOLOv5 network.

[0098] The vehicle-mounted camera transmits real-time road images to the vehicle controller; the vehicle controller uses the pedestrian classification model to classify and detect pedestrians in the real-time road images to obtain pedestrian classification results; the vehicle-mounted sensor detects the distance between pedestrians and the vehicle and obtains the current vehicle status, and transmits this information to the vehicle controller; the vehicle controller predicts whether a collision between a pedestrian and the vehicle will occur based on the distance and the current vehicle status; if a collision is predicted, the vehicle's active energy absorption device deploys the corresponding adult or child airbag according to the pedestrian classification results; otherwise, monitoring continues.

[0099] It should be noted that the above embodiments are described from the perspective of hardware devices to illustrate the technical solutions of the present invention.

[0100] In some embodiments, the present invention also provides a computer-readable storage medium storing a computer program, which is invoked by a processor to implement the following:

[0101] Acquire real-time images of the road;

[0102] The real-time road image is input into a pedestrian classification model built on a YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children.

[0103] The distance between the pedestrian and the car, as well as the current vehicle status, are obtained, and based on the distance and the current vehicle status, it is predicted whether a collision between the pedestrian and the car will occur.

[0104] If a collision is predicted, the vehicle's active energy absorption device deploys the corresponding adult or child airbag based on the pedestrian classification results; otherwise, monitoring continues.

[0105] Please refer to the explanation of the method above for the specific implementation process of each step.

[0106] The readable storage medium is a computer-readable storage medium, which can be an internal storage unit of the controller described in any of the foregoing embodiments, such as the controller's hard drive or memory. The readable storage medium can also be an external storage device of the controller, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the controller. Further, the readable storage medium can include both the controller's internal storage unit and external storage devices. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium can also be used to temporarily store data that has been output or will be output.

[0107] Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] It should be emphasized that the examples described in this invention are illustrative rather than limiting. Therefore, this invention is not limited to the examples described in the specific embodiments. Any other embodiments derived by those skilled in the art based on the technical solutions of this invention, without departing from the spirit and scope of this invention, whether modifications or substitutions, are also within the protection scope of this invention.

Claims

1. A control method for an active energy-absorbing device for automobiles based on pedestrian classification and recognition, characterized in that: An active energy absorption device for automobiles is installed at the front of the vehicle and is an adult / child airbag structure, wherein the method includes: Real-time images of the road are obtained by using a vehicle-mounted camera; The real-time road image is input into a pedestrian classification model built on a YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children. The distance between pedestrians and cars is detected by using vehicle-mounted sensing devices and the current vehicle status is obtained. Based on the distance between pedestrians and cars and the current vehicle status, it is predicted whether a collision between pedestrians and cars will occur. If a collision is predicted, the vehicle's active energy absorption device will deploy the corresponding adult or child airbag based on the pedestrian classification results; otherwise, monitoring will continue. The YOLOv5 network comprises an input terminal, a backbone network, a neck network, and an output terminal connected in sequence. The backbone network includes a Focus layer, two convolutional layers (CBS), a first C3 module, a convolutional layer (CBS), a second C3 module, a convolutional layer (CBS), a third C3 module, a convolutional layer (CBS), a CA attention module, a spatial pyramid pooling module (SPPC), and a fourth C3 module, all connected in sequence. In this system, the first C3 module leads out another output path, and the second C3 module on the backbone network leads out another output path. The two paths are merged, and then pass through the Focus layer, the CBS convolutional layer, and the feature fusion module in sequence before being connected to the traditional transmission path from the second C3 module to the neck network, in order to improve the feature capture capability of children's targets. The Focus layer implements a slicing operation, taking the value of the input image every other pixel, thereby obtaining 4 independent feature layers, which expands the RGB 3 channels of the original image by 4 times, while reducing the height and width of the original image by 1 / 2. The relationship between the gas volume deployed by the adult or child airbag and the head shape acceleration and head shape mass is as follows: ; In the formula, P is the gas pressure; V is the gas volume; n is the gas quantity; R is the gas constant; T is the gas temperature; m is the head mass; and S is the area of ​​the head in the direction of the velocity vector. a This refers to head shape acceleration.

2. The method according to claim 1, characterized in that: The SPP module in the traditional backbone network is replaced with Spatial Pyramid Pooling (SPPCSPC) to enhance network robustness and improve the anti-interference level of child target recognition. The SPPCSPC is divided into two transmission paths. The first transmission path has three convolutional layers (CBS), a max pooling layer, a feature fusion module, and two convolutional layers (CBS) in sequence. The second transmission path has a convolutional layer (CBS). The output of the last convolutional layer (CBS) in the first transmission path is fused with the output of the last convolutional layer (CBS) in the second transmission path, and then input into the convolutional layer (CBS).

3. The method according to claim 1, characterized in that: The current vehicle status includes at least the current vehicle speed. The process of predicting whether a collision between a pedestrian and a car will occur based on the distance and the current vehicle status is as follows: If the pedestrian is an adult and the current vehicle speed exceeds a preset vehicle speed threshold V1 and the distance is less than a preset distance threshold L1, it is determined that the adult and the car will collide. If the pedestrian is a child and the current vehicle speed exceeds a preset vehicle speed threshold V2 and the distance is less than a preset distance threshold L2, it is determined that the child and the car will collide. The preset vehicle speed threshold V1 and the preset vehicle speed threshold V2 are in the range of 5.6 m / s to 11.1 m / s. The preset distance threshold L1 and the preset distance threshold L2 are in the range of 6.7m to 13.3m.

4. The method according to claim 1, characterized in that: Based on the aforementioned relationship and the approximately equivalent head mass and head acceleration of adults and children, the gas volume of the corresponding adult-type airbag and child-type airbag is determined. The uniform pressure method was used to simulate the active energy absorption device of the car, and the volume of the airbag at each moment during deployment met the following requirements: ; In the formula, The pressure of the airbag; This refers to the volume of the airbag; The mass of the gas inside the airbag; The temperature of the gas inside the airbag; The state constant of an ideal gas; The density of the gas; It is the energy of the gas; It is the heat capacity ratio constant; The rate of change of the mass of the gas inside the volume is controlled by t The mass of the gas flowing across the boundary at any given moment is determined by: ; In the formula, The mass flow rate of gas flowing into the airbag is the mass flow rate of gas from the gas generator into the airbag. ; The mass flow rate of gas exiting the airbag, i.e., the mass flow rate exiting through the exhaust port. and leaked mass flow sum; Based on the above formula, the state parameters of volume and pressure at each moment during the airbag deployment process are determined, thereby simulating and controlling the airbag deployment volume or the amount of gas in the airbag, and approximately controlling the peak acceleration of the head-shaped impact airbag.

5. A control system based on the method of any one of claims 1-4, characterized in that: include: The image acquisition module is used to acquire real-time images of the road. The pedestrian classification module is used to input the real-time road image into a pedestrian classification model built based on the YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children. The collision detection module is used to obtain the distance between the pedestrian and the car and the current vehicle status, and to predict whether a collision between the pedestrian and the car will occur based on the distance between the pedestrian and the car and the current vehicle status. The control module is used to deploy the corresponding adult or child airbags from the vehicle's active energy absorption device if a collision is predicted, based on the pedestrian classification results; otherwise, monitoring continues.

6. An active energy absorption device for automobiles based on the method of any one of claims 1-4, characterized in that: The active energy absorption device for automobiles includes child-type airbags and adult-type airbags, or the active energy absorption device for automobiles includes airbags and airbag inflation equipment, wherein the airbag inflation equipment controls the gas volume inside the airbag according to the child or adult type to obtain a child-type airbag or an adult airbag.

7. A vehicle-mounted system based on the method of any one of claims 1-4, characterized in that: At least including: Automotive controllers, automotive active energy absorption devices, vehicle cameras, and vehicle sensing equipment; Among them, the vehicle active energy absorption device, vehicle camera, and vehicle sensing device are all connected to the vehicle controller, and the vehicle controller calls or loads the pedestrian classification model built based on the YOLOv5 network. The vehicle-mounted camera transmits real-time images of the road it captures to the vehicle controller. The vehicle controller uses the pedestrian classification model to classify and detect the real-time road image to obtain the pedestrian classification result. The vehicle-mounted sensing device detects the distance between pedestrians and the vehicle and obtains the current vehicle status, and transmits it to the vehicle controller; The vehicle controller predicts whether a collision will occur based on the distance between the pedestrian and the vehicle and the current vehicle status. If a collision is predicted, the vehicle's active energy absorption device deploys the corresponding adult or child airbag according to the pedestrian classification result; otherwise, monitoring continues.

8. A computer-readable storage medium, characterized in that: The computer program is stored and is invoked by the processor to implement: Acquire real-time images of the road; The real-time road image is input into a pedestrian classification model built on a YOLOv5 network for classification and detection to obtain pedestrian classification results. The pedestrian classification results are used to identify whether there are pedestrians in the real-time road image and whether the pedestrians are adults or children. The distance between the pedestrian and the car, as well as the current vehicle status, are obtained, and based on the distance and the current vehicle status, it is predicted whether a collision between the pedestrian and the car will occur. If a collision is predicted, the vehicle's active energy absorption device will deploy the corresponding adult or child airbag based on the pedestrian classification results; otherwise, monitoring will continue. The YOLOv5 network comprises an input terminal, a backbone network, a neck network, and an output terminal connected in sequence. The backbone network includes a Focus layer, two convolutional layers (CBS), a first C3 module, a convolutional layer (CBS), a second C3 module, a convolutional layer (CBS), a third C3 module, a convolutional layer (CBS), a CA attention module, a spatial pyramid pooling module (SPPC), and a fourth C3 module, all connected in sequence. In this system, the first C3 module leads out another output path, and the second C3 module on the backbone network leads out another output path. The two paths are merged, and then pass through the Focus layer, the CBS convolutional layer, and the feature fusion module in sequence before being connected to the traditional transmission path from the second C3 module to the neck network, in order to improve the feature capture capability of children's targets. The Focus layer implements a slicing operation, taking the value of the input image every other pixel, thereby obtaining 4 independent feature layers, which expands the RGB 3 channels of the original image by 4 times, while reducing the height and width of the original image by 1 / 2. The relationship between the gas volume deployed by the adult or child airbag and the head shape acceleration and head shape mass is as follows: ; In the formula, P is the gas pressure; V is the gas volume; n is the gas quantity; R is the gas constant; T is the gas temperature; m is the head mass; and S is the area of ​​the head in the direction of the velocity vector. a This refers to head shape acceleration.

Images