Vehicle automatic emergency braking control method, device, equipment and medium

Abstract

Embodiments of the present application provide a vehicle automatic emergency braking control method, device, equipment and medium, wherein the method comprises: collecting first perception data of the surrounding environment of the vehicle through a low-cost laser radar, collecting second perception data of the surrounding environment of the vehicle through a low-resolution millimeter wave radar, and performing space-time synchronization on the first perception data and the second perception data; dynamically determining a first weight of the low-cost laser radar and a second weight of the low-resolution millimeter wave radar according to different situations of a target; weighting and fusing the first perception data and the second perception data according to the first weight and the second weight to obtain multi-modal data; calculating relative driving state information of the vehicle and the target based on the multi-modal data using a lightweight algorithm model, and triggering hierarchical braking according to different relative driving state information. The scheme reduces hardware cost and system complexity, improves environmental adaptability and perception accuracy, and improves the real-time performance of the AEB system.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and in particular to a control method, device, equipment and medium for automatic emergency braking of a vehicle. Background Technology

[0002] Automatic Emergency Braking (AEB) is a core function of active safety in intelligent driving, and its technological evolution has progressed from single-sensor to multi-modal fusion. The core challenges of existing AEB systems include:

[0003] The contradiction between cost and complexity: LiDAR is used in high-end vehicles (such as the Li Auto AD Max system) due to its high precision, but its high cost (the cost of a single-line LiDAR is about 500, but the cost of a multi-line LiDAR (such as 128-line) exceeds 2000) makes it difficult to popularize in the low-end and mid-range markets;

[0004] Insufficient environmental adaptability: Pure vision solutions (such as Tesla's early Autopilot) are greatly affected by lighting and rain / fog. Although millimeter-wave radar has strong anti-interference capabilities, low-resolution versions (such as 24GHz radar) have poor ability to identify static targets.

[0005] Real-time bottleneck: Traditional fusion algorithms (such as target-level fusion) need to process heterogeneous data from multiple sensors, resulting in response delays (typically >150ms), which makes it difficult to meet the needs of emergency braking.

[0006] The existing automatic emergency braking implementation schemes are as follows:

[0007] Option 1: Camera + Millimeter-wave Radar Fusion System

[0008] Structural principle: It adopts a combination of 77GHz millimeter-wave radar (detection range 150-250m) and monocular / dual-eye camera (resolution 1280×720), and uses target-level fusion (such as Hungarian algorithm) to associate radar range / velocity data with camera semantic information.

[0009] Representative products:

[0010] Bosch's third-generation AEB system integrates forward-facing radar and camera, supporting pedestrian detection in urban conditions, but it relies on high-resolution radar (costing approximately $300) and has a high false trigger rate for static targets.

[0011] Volvo City Safety: It uses radar + camera fusion and TTC (Time-of-Collision) model for decision-making, but the false negative rate for low obstacles (such as barbed wire) exceeds 15%.

[0012] limitation:

[0013] High-resolution radar and cameras require complex calibration, with a system cost of approximately $500; low-resolution millimeter-wave radar (such as 24GHz) has insufficient accuracy (±0.5m) at close range (<30m), leading to false braking or missed detection; data fusion takes a long time (average 120ms).

[0014] Option 2: Fusion of low-cost lidar and millimeter-wave radar

[0015] Structural principle: It combines a multi-line lidar (0.1° angular resolution) with a low-resolution millimeter-wave radar (such as 79GHz) and improves target tracking accuracy by synchronizing point cloud with radar signal in time and space (such as Kalman filtering).

[0016] Representative products:

[0017] Ideal Lidar Pilot: Integrates 128-line lidar + millimeter-wave radar to achieve braking to a stop at 80km / h in rain and fog, but the system cost exceeds $1500 and has high power consumption (>30W);

[0018] Mobileye EyeQ5: Supports raw LiDAR data fusion, but relies on a dedicated ASIC chip, resulting in poor hardware compatibility.

[0019] limitation:

[0020] Multi-line lidar is limited by cost and size constraints, making lightweight design difficult; single-line lidar has sparse point clouds (thousands of points per second), making target classification at long distances (>80m) challenging.

[0021] Option 3: Lightweight Multimodal Fusion Solution

[0022] Structural principle: Low-resolution millimeter-wave radar (such as 4D imaging radar) is fused with vision, and the model parameters are compressed (<10MB) through a lightweight neural network (such as MobileNet) to reduce computing power requirements.

[0023] Representative products:

[0024] Huawei MDC platform: Supports millimeter-wave radar + camera fusion with a response time of 110ms, but relies on a high-computing-power domain controller (power consumption > 20W);

[0025] Jimu Intelligent Commercial Vehicle AEB: Based on vision + radar fusion, the false trigger rate in backlight scenes is less than 5%, but without integrated LiDAR, the false detection rate of small targets at night is still relatively high.

[0026] limitation:

[0027] In dynamic scenarios, fixed sensor weights (such as reliance on radar for long distances) lead to errors in the velocity estimation of close-range targets; data preprocessing (such as point cloud filtering) consumes more than 30% of computing resources, resulting in significant latency bottlenecks.

[0028] Therefore, the existing technical solutions have the following drawbacks:

[0029] (1) High hardware cost and complex system

[0030] Multi-line LiDAR dependency: Existing solutions (such as Ideal Lidar Pilot) require a 128-line LiDAR (costing over $2000) to achieve high-precision point clouds, resulting in high system costs;

[0031] High-resolution sensor redundancy: Bosch, Volvo and other solutions rely on a combination of high-resolution millimeter-wave radar (77GHz, cost $300+) and cameras, which has high hardware redundancy and complex calibration (error must be <0.1°).

[0032] (2) Insufficient environmental adaptability and sensing accuracy

[0033] The limitations of single sensors: Pure vision solutions (early Tesla Autopilot) fail in backlight, rain, and fog; low-resolution millimeter-wave radar (24GHz) has a large ranging error (±0.5m) for static targets;

[0034] Rigid fusion strategy: Existing fusion systems (such as Huawei MDC) have fixed sensor weights, relying on radar for long distances and cameras for short distances, resulting in target speed estimation errors of >15% in dynamic scenarios (such as vehicle intrusion).

[0035] (3) Poor real-time performance and high computing power requirements

[0036] Data fusion latency: Traditional target-level fusion (such as Mobileye EyeQ5) requires multi-level data alignment and matching, with an average latency of 120-150ms;

[0037] Preprocessing resource consumption: Point cloud filtering and feature extraction consume more than 30% of computing power (such as the Jimu Intelligent Solution), resulting in a decision delay of more than 100ms.

[0038] (4) Lightweight design and insufficient energy efficiency

[0039] High power consumption of multi-line radar: Systems such as LiDAR Pilot consume more than 30W of power, requiring independent heat dissipation design, which makes them difficult to integrate into small vehicles;

[0040] Dedicated hardware dependency: Mobileye's solution requires ASIC chip support, resulting in poor hardware compatibility and high upgrade costs. Summary of the Invention

[0041] In view of this, embodiments of the present invention provide a control method for automatic emergency braking of a vehicle, to solve the technical problems of high cost, system complexity, insufficient environmental adaptability and perception accuracy, and poor real-time performance in existing automatic emergency braking schemes. The method includes:

[0042] The vehicle acquires first perception data of its surrounding environment using a low-cost lidar and a second perception data of its surrounding environment using a low-resolution millimeter-wave radar. The first perception data and the second perception data are then spatiotemporally synchronized. The low-cost lidar may include a single-line lidar or a solid-state MEMS lidar.

[0043] The first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar are dynamically determined based on the different characteristics of the target.

[0044] The first perceived data and the second perceived data are weighted and fused according to the first weight and the second weight to obtain multimodal data;

[0045] Based on the multimodal data, a lightweight algorithm model is used to calculate the relative driving state information between the vehicle and the target, and the braking is triggered in stages according to the different relative driving state information.

[0046] This invention also provides a control device for automatic emergency braking of vehicles, to solve the technical problems of high cost, system complexity, insufficient environmental adaptability and perception accuracy, and poor real-time performance in existing automatic emergency braking schemes. The device includes:

[0047] The first data acquisition module is used to acquire first perception data of the environment around the vehicle through a low-cost lidar, acquire second perception data of the environment around the vehicle through a low-resolution millimeter-wave radar, and perform spatiotemporal synchronization of the first perception data and the second perception data. The low-cost lidar includes a single-line lidar or a solid-state MEMS lidar.

[0048] The weight determination module is used to dynamically determine the first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar according to different target conditions.

[0049] The data fusion module is used to perform weighted fusion of the first perceived data and the second perceived data according to the first weight and the second weight to obtain multimodal data;

[0050] The braking triggering module is used to calculate the relative driving state information between the vehicle and the target based on the multimodal data using a lightweight algorithm model, and to trigger graded braking according to the different relative driving state information.

[0051] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned automatic emergency braking control methods for vehicles, thereby solving the technical problems of high cost, system complexity, insufficient environmental adaptability and perception accuracy, and poor real-time performance of existing automatic emergency braking schemes.

[0052] This invention also provides a computer-readable storage medium storing a computer program that executes any of the above-described automatic emergency braking control methods for vehicles, in order to solve the technical problems of high cost, system complexity, insufficient environmental adaptability and perception accuracy, and poor real-time performance of existing automatic emergency braking schemes.

[0053] Compared with the prior art, the beneficial effects that the above-mentioned at least one technical solution adopted in the embodiments of this specification can achieve include at least the following: A combination of low-cost LiDAR and low-resolution millimeter-wave radar is proposed. The low-cost LiDAR can achieve low cost and lightweight design, while the low-resolution millimeter-wave radar has strong anti-interference characteristics. This combination can reduce hardware costs and system complexity while ensuring basic sensing capabilities, which is conducive to achieving lightweight design and solves the problems of high complexity and cost of existing multi-sensor fusion AEB systems. Simultaneously, by dynamically determining the first weight of the low-cost LiDAR and the second weight of the low-resolution millimeter-wave radar according to different target conditions, the first sensing capability is achieved. The known data and the second sensing data are weighted and fused based on dynamic weights to achieve full complementarity of multimodal data. This allows the all-weather characteristics of low-resolution millimeter-wave radar to compensate for the weather sensitivity of lidar, while the high-precision point cloud data of lidar improves the accuracy of millimeter-wave radar in close-range target identification. This, in turn, helps to improve environmental adaptability and sensing accuracy, and solves the problem of insufficient sensing reliability in severe weather and complex environments. In addition, a lightweight algorithm model is used based on the multimodal data to calculate the relative driving state information between the vehicle and the target and trigger graded braking, which helps to shorten the response time, achieve rapid response, and improve the real-time performance of the AEB system. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of a control method for automatic emergency braking of a vehicle provided in an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the principle framework of a control method for implementing automatic emergency braking of a vehicle provided in an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of a lightweight algorithm model provided in an embodiment of the present invention;

[0058] Figure 4 This is a structural block diagram of a computer device provided in an embodiment of the present invention;

[0059] Figure 5 This is a structural block diagram of a vehicle automatic emergency braking control device provided in an embodiment of the present invention. Detailed Implementation

[0060] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0061] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] In this embodiment of the invention, a control method for automatic emergency braking of a vehicle is provided, such as... Figure 1 As shown, the method includes:

[0063] S101: First perception data of the environment around the vehicle is collected by a low-cost lidar, and second perception data of the environment around the vehicle is collected by a low-resolution milli-wave radar. The first perception data and the second perception data are spatiotemporally synchronized. The low-cost lidar includes a single-line lidar or a solid-state MEMS lidar.

[0064] S102: Dynamically determine the first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar according to the different situations of the target;

[0065] S103: The first perceived data and the second perceived data are weighted and fused according to the first weight and the second weight to obtain multimodal data;

[0066] S104: Based on the multimodal data, a lightweight algorithm model is used to calculate the relative driving state information between the vehicle and the target, and the graded braking is triggered according to the different relative driving state information.

[0067] Depend on Figure 1 As shown in the flowchart, this embodiment of the invention proposes a combination of low-cost LiDAR and low-resolution millimeter-wave radar. The low-cost LiDAR achieves low cost and lightweight design, while the low-resolution millimeter-wave radar possesses strong anti-interference capabilities. This combination reduces hardware costs and system complexity while ensuring basic perception capabilities, facilitating lightweight design and addressing the high complexity and cost issues of existing multi-sensor fusion AEB systems. Simultaneously, by dynamically determining the first weight of the low-cost LiDAR and the second weight of the low-resolution millimeter-wave radar based on different target conditions, the first and second perception data are weighted and fused based on dynamic weights, achieving full complementarity of multimodal data. This allows the all-weather characteristics of the low-resolution millimeter-wave radar to compensate for the weather sensitivity of the LiDAR, while the high-precision point cloud data of the LiDAR improves the accuracy of the millimeter-wave radar in near-range target identification, thereby improving environmental adaptability and perception accuracy and addressing the problem of insufficient perception reliability in adverse weather and complex environments. Furthermore, based on the multimodal data, a lightweight algorithm model is used to calculate the relative driving state information between the vehicle and the target for graded braking triggering, which helps to shorten response time, achieve rapid response, and improve the real-time performance of the AEB system.

[0068] In practical implementation, a combination of single-line LiDAR (cost <500) and low-resolution millimeter-wave radar (24GHz / 79GHz, cost <100) is used to replace multi-line LiDAR (such as 128-line radar $2000+) and high-resolution radar, reducing hardware costs by more than 60%. Simultaneously, the calibration process is simplified; spatiotemporal synchronization of the first and second sensing data allows multi-sensor calibration errors to be controlled within ±0.2°, reducing deployment complexity. The combination of single-line LiDAR (volume <0.5L) and low-resolution millimeter-wave radar (power consumption <5W) results in a total system power consumption of <15W, making it suitable for compact vehicles.

[0069] In practical implementation, a single-line lidar can be a mechanically rotating single-line lidar (such as the Hesai PandarXT, with a cost of <$500). It acquires a two-dimensional point cloud through horizontal scanning (e.g., a horizontal field of view of 120°, a vertical field of view of 3°, and an angular resolution of 0.1°) (maximum detection range of 150m), with only 3° vertical coverage to simplify data processing. The single-line lidar has a volume of <0.5L and power consumption of <8W, making it suitable for installation in compact vehicles. A low-resolution millimeter-wave radar can be a 24GHz radar (such as the Continental ARS540, with a cost of <$80), with a detection range of 80m and a resolution of 5°. This radar itself can improve velocity measurement accuracy (±0.1m / s) through Doppler frequency shift compensation. The millimeter-wave radar can also be upgraded to a 79GHz radar to achieve improved rain and fog penetration capabilities at a low cost.

[0070] In practical implementation, a combination of solid-state MEMS LiDAR (such as InnovizOne, which has a similar cost to single-line LiDAR) and low-resolution millimeter-wave radar (24GHz / 79GHz, with a cost of <100) can be used to replace multi-line LiDAR (such as 128-line radar $2000+) and high-resolution radar, thereby reducing hardware costs and deployment complexity.

[0071] In specific implementation, such as Figure 2 As shown, taking a single-line lidar as an example, the low-cost lidar acquires the first sensing data (such as two-dimensional point cloud data) and then preprocesses it. The low-resolution millimeter-wave lidar acquires the second sensing data (such as a velocity matrix) and then preprocesses it as well. To further improve response speed, parallel preprocessing acceleration is proposed. For example, distance-based adaptive voxel filtering is performed on the first sensing data to optimize point cloud filtering. The voxel size is 0.1m for near-range and 0.5m for far-range, reducing computation by 60%. Simultaneously, Doppler compensation and static target filtering are performed in parallel on the second sensing data to achieve millimeter-wave lidar data correction. This can be accelerated by a GPU, reducing the processing time to <10ms.

[0072] In specific implementation, during the data preprocessing stage, during the spatiotemporal synchronization of the preprocessed first and second sensing data, to further simplify the calibration process, it is proposed that a synchronization pulse signal can be generated using an FPGA to align the timestamps of the lidar and millimeter-wave radar data (i.e., the first and second sensing data) with an error of <1ms. Alternatively, software interpolation can be used to align the timestamps of the first and second sensing data. Spatial alignment of the first and second sensing data can be achieved by converting the radar polar coordinate data into the lidar coordinate system based on the extrinsic parameter calibration matrix (calibration error <±0.2°).

[0073] In specific implementation, such as Figure 2 As shown, after preprocessing the first and second sensing data, they can be fused. To improve data accuracy and environmental adaptability, a weighted fusion of the first and second sensing data based on dynamic weights is proposed. The first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar are dynamically determined according to different target conditions, including:

[0074] If the target is a vehicle, then different first weights and second weights are determined based on the different distances between the vehicle and the target;

[0075] If the target is a type of target other than a vehicle, then the first weight is increased and the first weight is greater than the second weight.

[0076] Specifically, if the target is a vehicle, different first weights and second weights are determined based on the different distances between the vehicle and the target, and the sum of the first weights and the second weights is 100%, including:

[0077] If the distance between the vehicle and the target is less than a first distance threshold (e.g., 30m), then the first weight is determined to be greater than or equal to 70%, and the second weight is determined to be less than 30%. In this case, the high-precision point cloud (i.e., the first perception data) is relied upon, and the radar's second perception data is only used to compensate for speed information.

[0078] If the distance between the vehicle and the target is greater than or equal to the first distance threshold and less than or equal to the second distance threshold (e.g., 80m), then it is determined that the values ​​of the first weight and the second weight are close and the second weight is greater than the first weight (e.g., the first weight is 40% and the second weight is 60%), wherein the second distance threshold is greater than the first distance threshold.

[0079] If the distance between the vehicle and the target is greater than the second distance threshold, then the second weight is determined to be greater than or equal to 90%; if the second weight is determined to be less than 10%, then the lidar is only used for target verification.

[0080] If the target is a type of target other than a vehicle, such as a pedestrian, obstacle, or cyclist, then the first weight is increased and the first weight is greater than the second weight, such as the first weight being 85%.

[0081] In specific implementation, during the data fusion process, in order to reduce data fusion latency, improve response speed, and enhance system lightweightness, a lightweight algorithm model is proposed to calculate the relative driving state information between the vehicle and the target based on the multimodal data, including:

[0082] The lightweight neural network model is compressed to obtain a compressed neural network model. The multimodal data is input into the compressed neural network model, and the relative driving state information is calculated through the compressed neural network model. The lightweight neural network model is either a backbone network or a YOLO-Fastest.

[0083] In specific implementation, such as Figure 3 As shown, the lightweight neural network model, taking a backbone network (such as MobileNetV3) as an example, can also use YOLO-Fastest. The lightweight neural network model is compressed using model compression technology (channel pruning, parameter reduction of 40%, and 8-bit quantization, model size <5MB) to reduce model parameters, resulting in a compressed neural network model. Multimodal data is then input into the compressed neural network model for inference and computation to obtain the relative driving state information between the vehicle and the target, improving inference speed. For example, the compressed neural network model running on the Jetson Nano platform can achieve an inference speed of 25 FPS (latency <40ms), compressing the total system response time from 150ms in the traditional solution to 90ms. Simultaneously, resource consumption is optimized, the computational load of the fusion algorithm is reduced by 50%, it can run on low-cost embedded platforms (such as Jetson Nano), is compatible with general hardware, and the algorithm supports mainstream automotive computing platforms (such as TITDA4), requiring no dedicated ASIC chip.

[0084] In specific implementation, such as Figure 2 As shown, after obtaining relative driving state information through multimodal data fusion, a braking strategy is executed. To improve the accuracy and smoothness of braking triggering, a graded braking triggering based on different relative driving state information is proposed. For example, if the combined confidence level of the low-cost lidar and the low-resolution millimeter-wave radar is greater than or equal to a preset confidence threshold, then the corresponding braking mechanism is triggered according to the different risk levels of the relative driving state information.

[0085] If the combined confidence level of the low-cost lidar and the low-resolution millimeter-wave radar is less than a preset confidence threshold (e.g., a combined confidence level of 60-85%), then the auxiliary data is used to verify whether the target exists in a preset number of consecutive frames. If so, the corresponding braking mechanism is triggered based on the different risk levels of the relative driving state information. The auxiliary data includes the vehicle's historical predicted trajectory (e.g., Kalman filtering) or the vehicle's camera data.

[0086] In specific implementation, the relative driving state information can be TTC (Time to Collision between the host vehicle and the target) or PDR (Remaining distance between the host vehicle and the target). The risk levels (such as high risk, medium risk, low risk) can be divided respectively according to the numerical values of TTC or PDR, and then different braking mechanisms can be triggered based on different risk levels. For example, taking TTC as an example,

[0087] When the comprehensive confidence level of the sensor (lidar + millimeter-wave radar) > 85% (i.e., the preset confidence threshold), the first TTC threshold is 1.5 s, and the second TTC threshold is 3 s. Different braking mechanisms are triggered based on different risk levels as follows:

[0088] High risk (TTC < 1.5 s): Trigger full braking (deceleration > 6 m / s 2 );

[0089] Medium risk (1.5 s < TTC < 3 s): Trigger pre-braking (deceleration is 2 - 4 m / s 2 ) and issue an audible and visual alarm;

[0090] Low risk (TTC > 3 s): Only give a warning without braking.

[0091] In specific implementation, the above-mentioned comprehensive confidence level can be the weighted sum or the mean of the confidence levels of a low-cost lidar and a low-resolution millimeter-wave radar.

[0092] In specific implementation, for the braking mechanism triggered by grading braking according to different relative driving state information, the false triggering rate is about 5% compared with the traditional scheme, and this application can reduce the false triggering rate to < 1%; By grading braking, the discomfort of passengers can be reduced, and the acceleration fluctuation < 0.5 m / s 3 .

[0093] In specific implementation, as shown in Table 1 below, this application has many beneficial effects compared with the prior art:

[0094] 1. Hardware cost reduction and lightweight design

[0095] Compared with the prior art: The traditional scheme relies on high-cost multi-line lidars (such as the cost of a 128-line radar exceeds 2000) or high-resolution millimeter-wave radars (such as a Bosch 77 GHz radar at 2000) or high-resolution millimeter-wave radars (such as a Bosch 77 GHz radar at 2000) or high-resolution millimeter-wave radars (such as a Bosch 77 GHz radar at 300+).

[0096] Advantages of the present invention:

[0097] By using single-line lidar (such as Hesai PandarXT, cost <500) and low-resolution millimeter-wave radar (such as 24GHz radar <500) and low-resolution millimeter-wave radar (such as 24GHz radar <500) and low-resolution millimeter-wave radar (such as 24GHz radar <100), the hardware cost is reduced by more than 60%.

[0098] The single-line lidar has a volume of <0.5L and a power consumption of <8W, while the millimeter-wave lidar has a power consumption of <5W. The total system power consumption is <15W, making it suitable for compact vehicles.

[0099] Technical support: By simplifying sensor configuration and optimizing space occupation, a breakthrough in both low cost and lightweight design is achieved.

[0100] 2. Improved environmental adaptability and sensing accuracy

[0101] Compared with existing technologies: the pure vision solution (Tesla's early Autopilot) fails in backlight / rain and fog, and the low-resolution millimeter-wave radar (24GHz) has a ranging error of ±0.5m.

[0102] Advantages of this invention:

[0103] Multimodal complementarity: Millimeter-wave radar compensates for the failure of lidar in rain and fog, and lidar corrects the short-range error of millimeter-wave radar (range measurement accuracy ±0.1m within 30m);

[0104] Dynamic weight allocation: Adjust sensor weights based on target distance (e.g., increase millimeter-wave radar weight to 80% in rainy or foggy weather) to improve robustness in complex scenarios.

[0105] Technical support: The fusion strategy breaks through the limitations of single sensors and achieves high-precision perception in all weather conditions.

[0106] 3. Optimization of response speed and computing efficiency

[0107] Compared with existing technologies: traditional fusion algorithms (such as Mobileye EyeQ5) have a response latency of 120-150ms and preprocessing computing power consumption of >30%.

[0108] Advantages of this invention:

[0109] Fast response algorithm: Lightweight neural network (MobileNetV3 compressed to <5MB) and parallel processing (GPU accelerated point cloud filtering), system response time ≤90ms;

[0110] Reduced computing power requirements: peak computing power <2 TOPS, can run on low-cost embedded platforms such as Jetson Nano.

[0111] Technical support: Algorithm optimization breaks through the real-time bottleneck, meeting the millisecond-level decision-making requirements of emergency braking.

[0112] 4. Dynamic decision-making mechanism and reduced false trigger rate

[0113] Compared to existing technologies, Bosch's AEB system has a false trigger rate of about 5% and cannot trigger braking in high-speed scenarios (>85km / h).

[0114] Advantages of this invention:

[0115] Tiered TTC thresholds: High risk (TTC < 1.5s) full braking, medium risk pre-braking, low risk only warning;

[0116] Confidence fusion decision: Direct triggering when the confidence level of dual sensors is >85%, and secondary verification by combining trajectory prediction with 60-85% confidence level, with a false trigger rate of <1%.

[0117] Technical support: A multi-dimensional verification mechanism enhances braking safety and passenger comfort.

[0118] 5. Technology compatibility and scalability

[0119] Compared to existing technologies, Bosch's solution relies on dedicated ASIC chips, resulting in poor hardware compatibility.

[0120] Advantages of this invention:

[0121] It supports mainstream automotive computing platforms (such as TI TDA4) and requires no dedicated chip.

[0122] It can be expanded to include an integrated camera as a third verification source, or upgraded to a 79GHz radar to enhance rain and fog penetration capabilities.

[0123] Technical support: The open architecture design leaves room for feature upgrades.

[0124] Table 1

[0125]

[0126] In practical implementation, the aforementioned automatic emergency braking control method for vehicles can also collect data through the following hardware combinations to meet the braking requirements of different application scenarios, for example...

[0127] In high-speed scenarios (such as commercial vehicles or mid-range passenger vehicles suitable for high-speed scenarios (>100km / h)), the first perception data of the vehicle's surrounding environment can be collected, and the second perception data of the vehicle's surrounding environment can be collected through a low-resolution camera (2 megapixels). The low-resolution camera performs target classification, and the speed measurement accuracy is improved (±0.05m / s) through a Doppler compensation algorithm; or;

[0128] In scenarios involving nighttime roads or frequent animal activity, the first perception data can be collected using a far-infrared thermal imaging camera, while the second perception data of the vehicle's surrounding environment can be collected using a low-resolution millimeter-wave radar.

[0129] Specifically, the high-resolution 4D imaging radar can utilize the Continental ARS540, which boasts a vertical resolution of 2°, a detection range of 300m, and the ability to generate point cloud-level data. The unit price of this high-resolution 4D imaging radar is less than 200, representing a 60% cost reduction compared to lidar (single-line <500). Millimeter-wave radar is unaffected by rain or fog, and can replace lidar in adverse weather conditions, ensuring environmental adaptability.

[0130] Specifically, far-infrared (LWIR) thermal imaging cameras (such as Teledyne FLIR), with a detection range of 150m, identify biological targets (pedestrians, animals) through temperature differences. Combined with a 24GHz low-resolution millimeter-wave radar, the data from both employs a feature-level fusion strategy (thermal imaging provides target category, and radar provides motion parameters). This combination improves pedestrian detection success rate to 92% under 0.2 lux illumination (compared to only 65% ​​for traditional cameras), enhancing nighttime performance; it also improves false trigger control, as thermal imaging can distinguish between live and non-live targets, reducing false detections such as metal signs.

[0131] In this embodiment, a computer device is provided, such as... Figure 4 As shown, it includes a memory 401, a processor 402, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-described automatic emergency braking control methods for vehicles.

[0132] Specifically, the computer device can be a computer terminal, a server, or a similar computing device.

[0133] In this embodiment, a computer-readable storage medium is provided, which stores a computer program that performs any of the above-described automatic emergency braking control methods for vehicles.

[0134] Specifically, computer-readable storage media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable storage media does not include transient media, such as modulated data signals and carrier waves.

[0135] Based on the same inventive concept, this invention also provides a control device for automatic emergency braking of a vehicle, as described in the following embodiments. Since the principle of the control device for automatic emergency braking of a vehicle is similar to that of the control method for automatic emergency braking of a vehicle, the implementation of the control device for automatic emergency braking of a vehicle can refer to the implementation of the control method for automatic emergency braking of a vehicle, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0136] Figure 5 This is a structural block diagram of a vehicle automatic emergency braking control device according to an embodiment of the present invention, such as... Figure 5 As shown, it includes:

[0137] The first data acquisition module 501 is used to acquire first perception data of the environment around the vehicle through a low-cost lidar, acquire second perception data of the environment around the vehicle through a low-resolution millimeter-wave radar, and perform spatiotemporal synchronization of the first perception data and the second perception data. The low-cost lidar includes a single-line lidar or a solid-state MEMS lidar.

[0138] The weight determination module 502 is used to dynamically determine the first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar according to different target conditions.

[0139] Data fusion module 503 is used to perform weighted fusion of the first perceived data and the second perceived data according to the first weight and the second weight to obtain multimodal data;

[0140] The braking trigger module 504 is used to calculate the relative driving state information between the vehicle and the target based on the multimodal data using a lightweight algorithm model, and to trigger graded braking according to the different relative driving state information.

[0141] In one embodiment, the weight determination module is configured to determine different first weights and second weights based on the different distances between the vehicle and the target if the target is a vehicle; and to increase the first weight and make the first weight greater than the second weight if the target is a target other than a vehicle.

[0142] In one embodiment, the weight determination module is configured to: if the distance between the vehicle and the target is less than a first distance threshold, determine that the first weight is greater than or equal to 70% and the second weight is less than 30%, wherein the sum of the first weight and the second weight is 100%; if the distance between the vehicle and the target is greater than or equal to the first distance threshold and less than or equal to a second distance threshold, determine that the values ​​of the first weight and the second weight are close and the second weight is greater than the first weight, wherein the second distance threshold is greater than the first distance threshold; if the distance between the vehicle and the target is greater than the second distance threshold, determine that the second weight is greater than or equal to 90% and the second weight is less than 10%.

[0143] In one embodiment, the braking triggering module is used to perform model compression processing on the lightweight neural network model to obtain a compressed neural network model, input the multimodal data into the compressed neural network model, and calculate the relative driving state information through the compressed neural network model, wherein the lightweight neural network model is a backbone network or YOLO-Fastest.

[0144] In one embodiment, the above-mentioned apparatus further includes:

[0145] The preprocessing module is used to perform distance-based adaptive voxel filtering on the first sensing data before weighted fusion of the first sensing data and the second sensing data according to the first weight and the second weight, and simultaneously perform Doppler compensation and static target filtering on the second sensing data in parallel.

[0146] In one embodiment, the braking triggering module is configured to: if the combined confidence level of the low-cost lidar and the low-resolution millimeter-wave radar is greater than or equal to a preset confidence threshold, trigger a corresponding braking mechanism based on the different risk levels of the relative driving state information; if the combined confidence level of the low-cost lidar and the low-resolution millimeter-wave radar is less than the preset confidence threshold, verify whether the target exists in a preset number of consecutive frames based on auxiliary data; if so, trigger a corresponding braking mechanism based on the different risk levels of the relative driving state information, wherein the auxiliary data includes the vehicle's historical predicted trajectory or the vehicle's camera data.

[0147] In one embodiment, the above-mentioned apparatus further includes:

[0148] The second data acquisition module is used to acquire the first perception data of the vehicle's surrounding environment through a high-resolution 4D imaging radar and the second perception data of the vehicle's surrounding environment through a low-resolution camera in high-speed scenarios; or, in nighttime scenarios or scenarios where animals are frequently seen, to acquire the first perception data through a far-infrared thermal imaging camera.

[0149] The embodiments of this invention achieve the following technical effects: A combination of low-cost LiDAR and low-resolution millimeter-wave radar is proposed. The low-cost LiDAR enables cost-effectiveness and lightweight design, while the low-resolution millimeter-wave radar possesses strong anti-interference capabilities. This combination reduces hardware costs and system complexity while maintaining basic perception capabilities, facilitating lightweight design and addressing the high complexity and cost issues of existing multi-sensor fusion AEB systems. Simultaneously, by dynamically determining the first weight of the low-cost LiDAR and the second weight of the low-resolution millimeter-wave radar based on different target conditions, weighted fusion of the first and second perception data based on dynamic weights is achieved. This ensures full complementarity of multimodal data, utilizing the all-weather characteristics of the low-resolution millimeter-wave radar to compensate for the weather sensitivity of the LiDAR. Furthermore, the high-precision point cloud data from the LiDAR enhances the accuracy of the millimeter-wave radar in near-range target identification, thereby improving environmental adaptability and perception accuracy and addressing the problem of insufficient perception reliability in adverse weather and complex environments. In addition, a lightweight algorithm model is used to calculate the relative driving state information between the vehicle and the target based on the multimodal data for graded braking triggering, which helps to shorten response time, achieve rapid response, and improve the real-time performance of the AEB system.

[0150] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0151] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A control method for automatic emergency braking of a vehicle, characterized in that, include: The vehicle acquires first perception data of its surrounding environment using a low-cost lidar and a second perception data of its surrounding environment using a low-resolution millimeter-wave radar. The first perception data and the second perception data are then spatiotemporally synchronized. The low-cost lidar may include a single-line lidar or a solid-state MEMS lidar. The first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar are dynamically determined based on the different characteristics of the target. The first perceived data and the second perceived data are weighted and fused according to the first weight and the second weight to obtain multimodal data; Based on the multimodal data, a lightweight algorithm model is used to calculate the relative driving state information between the vehicle and the target, and the graded braking is triggered according to the different relative driving state information. The first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar are dynamically determined based on different target conditions, including: If the target is a vehicle, then different first weights and second weights are determined based on the different distances between the vehicle and the target; If the target is a type of target other than a vehicle, then the first weight is increased and the first weight is greater than the second weight.

2. The control method for automatic emergency braking of a vehicle as described in claim 1, characterized in that, If the target is a vehicle, then different first weights and second weights are determined based on the different distances between the vehicle and the target, including: If the distance between the vehicle and the target is less than a first distance threshold, then the first weight is determined to be greater than or equal to 70%, and the second weight is determined to be less than 30%, wherein the sum of the first weight and the second weight is 100%. If the distance between the vehicle and the target is greater than or equal to the first distance threshold and less than or equal to the second distance threshold, then it is determined that the values ​​of the first weight and the second weight are close and the second weight is greater than the first weight, wherein the second distance threshold is greater than the first distance threshold; If the distance between the vehicle and the target is greater than the second distance threshold, then the first weight is determined to be greater than or equal to 90%, and the second weight is determined to be less than 10%.

3. The control method for automatic emergency braking of a vehicle as described in claim 1, characterized in that, Based on the multimodal data, a lightweight algorithm model is used to calculate the relative driving state information between the vehicle and the target, including: The lightweight neural network model is compressed to obtain a compressed neural network model. The multimodal data is input into the compressed neural network model, and the relative driving state information is calculated through the compressed neural network model. The lightweight neural network model is either a backbone network or a YOLO-Fastest.

4. The control method for automatic emergency braking of a vehicle as described in any one of claims 1 to 3, characterized in that, Also includes: Before weighted fusion of the first and second sensing data according to the first and second weights, distance-based adaptive voxel filtering is performed on the first sensing data, while Doppler compensation and static target filtering are performed in parallel on the second sensing data.

5. The control method for automatic emergency braking of a vehicle as described in any one of claims 1 to 3, characterized in that, Triggering graded braking based on the different relative driving state information includes: If the combined confidence level of the low-cost lidar and the low-resolution millimeter-wave radar is greater than or equal to a preset confidence threshold, then the corresponding braking mechanism is triggered according to the different risk levels of the relative driving state information. If the combined confidence level of the low-cost lidar and the low-resolution millimeter-wave radar is less than a preset confidence threshold, then the auxiliary data is used to verify whether the target exists in a preset number of consecutive frames. If so, the corresponding braking mechanism is triggered according to the different risk levels of the relative driving state information. The auxiliary data includes the vehicle's historical predicted trajectory or the vehicle's camera data.

6. The control method for automatic emergency braking of a vehicle as described in any one of claims 1 to 3, characterized in that, Also includes: In high-speed scenarios, the first perception data of the vehicle's surrounding environment is acquired by a high-resolution 4D imaging radar, and the second perception data of the vehicle's surrounding environment is acquired by a low-resolution camera; or; In nighttime roads or scenarios where animals are frequently seen, the first sensing data is collected using a far-infrared thermal imaging camera.

7. A control device for automatic emergency braking of a vehicle, characterized in that, include: The first data acquisition module is used to acquire first perception data of the environment around the vehicle through a low-cost lidar, acquire second perception data of the environment around the vehicle through a low-resolution millimeter-wave radar, and perform spatiotemporal synchronization of the first perception data and the second perception data. The low-cost lidar includes a single-line lidar or a solid-state MEMS lidar. The weight determination module is used to dynamically determine the first weight of the low-cost lidar and the second weight of the low-resolution millimeter-wave radar according to different target conditions. The data fusion module is used to perform weighted fusion of the first perceived data and the second perceived data according to the first weight and the second weight to obtain multimodal data; The braking triggering module is used to calculate the relative driving state information between the vehicle and the target based on the multimodal data using a lightweight algorithm model, and to trigger graded braking according to the different relative driving state information. The weight determination module is used to determine different first weights and second weights based on the different distances between the vehicle and the target if the target is a vehicle. If the target is a type of target other than a vehicle, then the first weight is increased and the first weight is greater than the second weight.

8. A computer 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 computer program, it implements the control method for automatic emergency braking of a vehicle as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that performs the control method for automatic emergency braking of a vehicle according to any one of claims 1 to 6.

Images