Input adjustment method and apparatus for intelligent-driving sensing network, and device and medium

Abstract

The present application relates to the technical field of intelligent driving. Disclosed are an input adjustment method and apparatus for an intelligent-driving sensing network, and a device and a medium. In the present application, spatiotemporal parameters required for a vehicle to reach a target intersection are firstly acquired, such that the sparseness degree of traffic signal images can be flexibly adjusted accurately on the basis of a spatiotemporal relationship between the vehicle and the intersection, thereby laying a foundation for subsequent accurate recognition of a traffic signal type. This process does not need to rely on a high-precision map, thereby avoiding the problem of a traffic signal information lag caused by slow updating of the high-precision map. Furthermore, the mode of adjusting the sparseness degree on the basis of the spatiotemporal parameters does not rely on simply increasing the GPU computing power to improve the image resolution and the recognition accuracy. Instead, the images are optimized ingeniously and accurately on the basis of actual situations, such that requirements for traffic signal image recognition are met without increasing hardware costs, thereby solving the problem of high costs caused by an increase in the GPU computing power.

Description

Input adjustment methods, devices, equipment and media for intelligent driving perception networks Cross-references to related applications

[0001] This disclosure claims priority to Chinese Patent Application No. 2025101127421, filed with the Chinese Patent Office on January 24, 2025, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure relates to, but is not limited to, the field of intelligent driving technology, specifically to input adjustment methods, devices, equipment, and media for intelligent driving perception networks. Background Technology

[0003] In the field of intelligent driving, intelligent driving perception neural networks can be used to identify obstacles (such as vehicles, two-wheeled vehicles, pedestrians, etc.) and signal signs (such as traffic lights, taillights of the vehicle in front, speed limit signs, etc.). Summary of the Invention

[0004] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0005] In a first aspect, embodiments of this disclosure provide an input adjustment method for an intelligent driving perception network, the method comprising:

[0006] Obtain the spatiotemporal parameters required for the vehicle to reach the target intersection, wherein the target intersection is the intersection the vehicle is about to pass through, and the spatiotemporal parameters indicate the distance between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection;

[0007] The traffic light image corresponding to the target intersection is acquired by the vehicle, and the original sparsity of the traffic light image is adjusted according to the spatiotemporal parameters to obtain the adjusted traffic light image.

[0008] The current traffic light type at the target intersection is obtained by recognizing the adjusted traffic light image through an intelligent driving perception network.

[0009] Furthermore, adjusting the original sparsity of the traffic light image according to the spatiotemporal parameters to obtain the adjusted traffic light image includes:

[0010] If the spatiotemporal parameter is less than or equal to a first preset threshold, the traffic light image is adjusted from its original sparsity level to a first sparsity level to obtain an adjusted traffic light image, wherein the first sparsity level is less than the original sparsity level; or

[0011] If the spatiotemporal parameter is greater than or equal to the second preset threshold, the traffic light image is adjusted from the original sparsity to the second sparsity to obtain the adjusted traffic light image, wherein the second sparsity is greater than the original sparsity.

[0012] Wherein, the first preset threshold is less than the second preset threshold.

[0013] Furthermore, adjusting the traffic light image from its original sparsity level to a first sparsity level to obtain the adjusted traffic light image includes:

[0014] Obtain the mapping relationship between the preset threshold and the degree of sparsity;

[0015] Based on the mapping relationship, the first sparsity degree corresponding to the first preset threshold is determined, and the sparsity degree of the traffic light image is adjusted to the first sparsity degree to obtain the adjusted traffic light image.

[0016] Further, adjusting the sparsity of the traffic light image to the first sparsity level to obtain the adjusted traffic light image includes:

[0017] Obtain the illumination data of the road segment where the vehicle is located and the lane position corresponding to the vehicle;

[0018] Based on the illumination data and the lane position, determine the target area in the traffic light image that needs to be sparsified.

[0019] The sparsity of the target region in the traffic light image is adjusted to the first sparsity level to obtain the adjusted traffic light image.

[0020] Furthermore, adjusting the traffic light image from its original sparsity level to a first sparsity level to obtain the adjusted traffic light image includes:

[0021] Detect traffic condition data of the road segment where the vehicle is located and vehicle driving data of the vehicle;

[0022] Determine the degree of influence of the traffic condition data and the vehicle driving data on traffic light recognition, and obtain the adjustment ratio corresponding to the degree of influence;

[0023] Based on the adjustment ratio and the original sparsity level, a first sparsity level is calculated, and the sparsity level of the traffic light image is adjusted to the first sparsity level to obtain the adjusted traffic light image.

[0024] Furthermore, determining the degree of influence of the traffic condition data and the vehicle driving data on traffic light recognition includes:

[0025] Extract road segment environmental data and road segment traffic flow data from the traffic condition data;

[0026] Based on the traffic flow data of the aforementioned road section, predict the probability that the traffic lights will be blocked;

[0027] Based on the vehicle driving data and the road section environment data, the accuracy of the traffic light recognition is predicted;

[0028] The degree of influence is calculated based on the probability and the accuracy.

[0029] Furthermore, adjusting the traffic light image from its original sparsity level to a second sparsity level to obtain the adjusted traffic light image includes:

[0030] Obtain the mapping relationship between the preset threshold and the degree of sparsity;

[0031] Based on the mapping relationship, the second sparsity degree corresponding to the second preset threshold is determined, and the sparsity degree of the traffic light image is adjusted to the second sparsity degree to obtain the adjusted traffic light image.

[0032] Furthermore, the method also includes:

[0033] Compare the spatiotemporal parameters with the third preset threshold;

[0034] If the spatiotemporal parameter is less than the third preset threshold and greater than the first preset threshold, the vehicle is adjusted from its original driving speed to the target driving speed, and the vehicle is controlled to collect traffic light images, wherein the target driving speed is less than the original driving speed.

[0035] If the traffic light type identified based on the traffic light image is a passable type, then after the vehicle passes through the target intersection, the vehicle's speed will be adjusted from the target speed to its original speed.

[0036] Secondly, embodiments of this disclosure provide an input adjustment device for an intelligent driving perception network, the device comprising:

[0037] The acquisition module is used to acquire the spatiotemporal parameters required for the vehicle to reach the target intersection, wherein the target intersection is the intersection through which the vehicle is about to pass, and the spatiotemporal parameters indicate the distance between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection;

[0038] An adjustment module is used to acquire the traffic light image corresponding to the target intersection collected by the vehicle, and adjust the original sparsity of the traffic light image according to the spatiotemporal parameters to obtain the adjusted traffic light image.

[0039] The recognition module is used to identify the adjusted traffic light image through the intelligent driving perception network to obtain the traffic light type of the target intersection.

[0040] Thirdly, embodiments of this disclosure provide a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the method described in the first aspect or any corresponding embodiment.

[0041] Fourthly, embodiments of this disclosure provide a computer-readable storage medium storing computer instructions for causing a computer to perform the methods described in the first aspect or any corresponding embodiment.

[0042] The method provided in this disclosure has the following beneficial effects:

[0043] This disclosure first obtains the spatiotemporal parameters required for a vehicle to reach the target intersection. Based on the spatiotemporal relationship between the vehicle and the intersection, it flexibly adjusts the sparsity of the traffic light image, laying the foundation for accurate traffic light type identification. This process does not rely on high-precision maps, avoiding the problem of delayed traffic light information caused by slow high-precision map updates. Furthermore, the method of adjusting the sparsity based on spatiotemporal parameters does not rely on simply increasing GPU computing power to improve image resolution and recognition accuracy. Instead, it cleverly and precisely optimizes the image according to the actual situation, meeting the requirements for traffic light image recognition without increasing hardware costs, thus solving the high cost problem associated with increasing GPU computing power.

[0044] After reading and understanding the accompanying diagrams and detailed descriptions, the other aspects can be understood. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the specific embodiments of this disclosure, the accompanying drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0046] Figure 1 is a flowchart illustrating an input adjustment method for an intelligent driving perception network according to some embodiments of the present disclosure.

[0047] Figure 2 is a schematic diagram of another sparsity adjustment according to some embodiments of the present disclosure.

[0048] Figure 3 is a flowchart illustrating an input adjustment method for another intelligent driving perception network according to some embodiments of the present disclosure.

[0049] Figure 4 is a flowchart illustrating another intelligent driving perception network input adjustment method according to some embodiments of the present disclosure.

[0050] Figure 5 is a structural block diagram of an input adjustment device for an intelligent driving perception network according to an embodiment of the present disclosure.

[0051] Figure 6 is a schematic diagram of the hardware structure of a computer device according to an embodiment of this disclosure. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0053] According to an embodiment of this disclosure, an embodiment of an input adjustment method for an intelligent driving perception network is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system including a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0054] In the field of intelligent driving, intelligent driving perception neural networks can be used to identify obstacles (such as vehicles, two-wheeled vehicles, pedestrians, etc.) and signal signs (such as traffic lights, taillights of the vehicle in front, speed limit signs, etc.). Because the aforementioned obstacles are relatively large and can be identified with the assistance of millimeter-wave radar, the image resolution requirements can be relatively lower. However, traffic lights are small, and in complex scenarios such as glare and rain, higher image resolution is needed to ensure recognition accuracy. Furthermore, higher image resolution requires higher computing power.

[0055] To ensure accurate traffic light recognition and facilitate smooth vehicle passage in Navigation on Autopilot (NOA) scenarios, conventional solutions include utilizing traffic light information from high-definition maps, increasing graphics processing unit (GPU) computing power, or a combination of both. However, high-definition maps update slowly, significantly limiting their guidance for vehicle passage. Therefore, the industry commonly uses methods such as increasing computing power or a combination of both, which, while solving the problem to some extent, increases computing costs.

[0056] To address the issues of slow updates or high computing costs associated with using high-precision maps and / or increasing GPU computing power, this disclosure provides an input adjustment method for an intelligent driving perception network. In some embodiments, the input adjustment method for the intelligent driving perception network can be executed by a control unit in the vehicle or by a remote server communicatively connected to the vehicle. Figure 1 is a flowchart of the input adjustment method for an intelligent driving perception network according to an embodiment of this disclosure. As shown in Figure 1, the process includes the following steps S101-S103.

[0057] Step S101: Obtain the spatiotemporal parameters required for the vehicle to reach the target intersection, where the target intersection is the intersection the vehicle is about to pass through, and the spatiotemporal parameters indicate the distance between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection.

[0058] In this embodiment, the spatiotemporal parameter refers to an indicator that comprehensively considers the spatial distance between the vehicle and the target intersection, as well as the time required for the vehicle to reach the intersection. The spatiotemporal parameter is used to dynamically adjust the sparsity of the traffic light image and the vehicle's speed based on the relative position of the vehicle and the intersection and / or the time required for the vehicle to reach the target intersection. By acquiring the spatiotemporal parameter, the state and needs of the vehicle at different stages can be predicted, thereby enabling the perception and control of the vehicle's driving state. For example, when the vehicle is close to the intersection and about to arrive (spatiotemporal parameter is small), the traffic light image needs to be processed more finely to ensure accurate recognition, while the vehicle speed is appropriately reduced to ensure safe passage; while when the vehicle is far from the intersection (spatiotemporal parameter is large), a relatively low image resolution processing strategy can be adopted to save computing resources, while maintaining a higher vehicle speed to improve driving efficiency.

[0059] Step S102: Obtain the traffic light image corresponding to the target intersection collected by the vehicle, and adjust the original sparsity of the traffic light image according to the spatiotemporal parameters to obtain the adjusted traffic light image.

[0060] In this embodiment of the disclosure, while the vehicle is in motion, it uses an onboard camera to collect real-time image information of the target intersection ahead, including traffic light images. These images serve as the basic data source for subsequent traffic light recognition and processing. The camera's viewing angle and shooting range are determined based on the camera's installation position and angle on the vehicle to ensure that relevant information about the target intersection, including the status and position of the traffic lights, can be captured.

[0061] Furthermore, adjusting the image sparsity based on spatiotemporal parameters is necessary because image resolution requirements vary under different conditions. For example, when a vehicle approaches an intersection, a higher image resolution is needed for more accurate traffic light recognition; conversely, when the vehicle is far from the intersection, the image resolution can be appropriately reduced to conserve computing resources and improve processing speed. By adjusting the original sparsity of the traffic light image according to spatiotemporal parameters, system performance and resource utilization can be optimized while ensuring accurate traffic light recognition.

[0062] In this embodiment of the disclosure, the original sparsity of the traffic light image is adjusted according to spatiotemporal parameters to obtain the adjusted traffic light image, including the following cases:

[0063] Case 1: Compare the spatiotemporal parameters with the first preset threshold. If the spatiotemporal parameters are less than or equal to the first preset threshold, adjust the traffic light image from the original sparsity level to the first sparsity level to obtain the adjusted traffic light image. The first sparsity level is less than the original sparsity level.

[0064] It should be noted that the first preset threshold can be set to 40m or T1 time, used to define the relative positional relationship between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection. The spatiotemporal parameters comprehensively reflect the spatial distance and time required for the vehicle to reach the target intersection. Comparing them with the first preset threshold can determine the stage at which the vehicle approaches the intersection. If the distance in the spatiotemporal parameters is ≤40m, it indicates that the vehicle is spatially close to the intersection; if the time is ≤T1, it indicates that the vehicle will arrive in the near future.

[0065] Furthermore, the sparsity level determines the image resolution. During normal driving, the vehicle processes the acquired traffic light images at the original sparsity level to balance computational resources and image recognition requirements. When the vehicle approaches an intersection (spatiotemporal parameters ≤ a first preset threshold), to improve the accuracy of traffic light recognition, it needs to be adjusted to a first sparsity level, which is lower than the original sparsity level. This reduces the image sparsity level and increases image detail and resolution. The sparsity level is negatively correlated with image sharpness.

[0066] In some embodiments, adjusting the sparsity of an image is performed as follows: The original image is divided into multiple sub-blocks; for each sub-block, the pixel value at a preset position, the maximum pixel value, the minimum pixel value, or the average pixel value of the sub-block are determined as the target pixel value; the adjusted image is determined based on the target pixel value of each sub-block. For example, as shown in Figure 2, an 8*8 original image is divided into 16 2*2 sub-blocks; for each 2*2 sub-block, the pixel value at a preset upper-left position is determined as the target pixel value; the adjusted 4*4 image is determined based on the target pixel value of each 2*2 sub-block (the pixel value corresponding to the dark gray block in Figure 2). This completes the adjustment of the image sparsity. The size of the sub-blocks can be adjusted to change the sparsity; larger sub-blocks result in greater sparsity, and smaller sub-blocks result in less sparsity.

[0067] Generally, assuming the original image has a resolution of 5.94 million pixels and a size of 3300*1800, the minimum pixel resolution that the system can use normally is 3 million pixels. When the vehicle is in normal driving conditions, the image is usually compressed to a size of 2000*1500. In this process, for the horizontal dimension, the original 165 pixels can be averaged down to 100 pixels.

[0068] However, when dealing with traffic light scenarios, higher resolution traffic light images are desired for more accurate traffic light identification. In some embodiments, the image resolution can be synchronously increased to the highest level; for example, sparsity reduction can be canceled, restoring the image's pixel count from 3 million pixels to 5.94 million pixels. In some embodiments, the pixel resolution can be increased only for a specific target region (e.g., the upper right region) in the traffic light image; for example, a high-resolution, uncompressed original image can be acquired, and the image of the specific target region can be selected from it, or the sparsity of that specific target region can be reduced.

[0069] This adjustment process aims to flexibly change the sparsity (e.g., resolution) of the image according to different needs and scenarios, in order to meet the different processing requirements of traffic light images at different stages of vehicle travel. When the vehicle is far from the intersection, a higher sparsity can save computing resources, while when the vehicle is approaching the intersection, a lower sparsity or pixel enhancement of certain areas of the traffic light image can ensure better traffic light recognition.

[0070] In this embodiment of the disclosure, the traffic light image is adjusted from the original sparsity level to the first sparsity level to obtain the adjusted traffic light image, as shown in Figure 3, including the following steps A1-A2.

[0071] Step A1: Obtain the mapping relationship between the preset threshold and the degree of sparsity.

[0072] Specifically, in adjusting the sparsity of traffic light images using intelligent driving perception networks, preset thresholds are used to define the relative positional relationship between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection. The vehicle stores a pre-defined mapping relationship between different thresholds and sparsity levels. This mapping relationship is set based on different distances between the vehicle and the target intersection and / or different times required for the vehicle to reach the target intersection, as well as the different resolution requirements for traffic light image recognition. By establishing such a mapping relationship, the appropriate level of sparsity for the traffic light image can be accurately determined based on the vehicle's actual spatiotemporal parameters. For example, when the vehicle is close to the intersection (when the spatiotemporal parameters meet specific threshold conditions), a higher image resolution is needed for accurate traffic light recognition, resulting in a lower sparsity level; while when the vehicle is far from the intersection, the image resolution requirement is relatively lower, and a higher sparsity level can be achieved.

[0073] Step A2: Determine the first sparsity level corresponding to the first preset threshold based on the mapping relationship, and adjust the sparsity level of the traffic light image to the first sparsity level to obtain the adjusted traffic light image.

[0074] Specifically, after obtaining the mapping relationship between the preset threshold and the sparsity level, when it is determined that the vehicle's spatiotemporal parameters are less than or equal to the first preset threshold, the corresponding first sparsity level can be determined based on the mapping relationship. For example, if the mapping relationship indicates that when the vehicle is less than or equal to 40m from the intersection (the preset threshold), the image sparsity level needs to be adjusted to a specific lower level, i.e., the first sparsity level. The first sparsity level is determined for the case of a vehicle approaching the intersection, in order to meet the higher requirements for traffic light recognition accuracy.

[0075] In this embodiment of the disclosure, the sparsity of the traffic light image is adjusted to a first sparsity level to obtain an adjusted traffic light image, including the following steps A201-A203.

[0076] Step A201: Obtain the illumination data of the road segment where the vehicle is located and the lane position corresponding to the vehicle.

[0077] It should be noted that the lighting data for the road segment where the vehicle is located includes current environmental lighting information, such as light intensity and direction. Different lighting conditions will affect the quality of traffic light images. The display effect of traffic lights in images will vary under different lighting conditions, such as daytime, nighttime, cloudy, sunny, backlighting, or front lighting. For example, in backlighting conditions, traffic lights may appear too bright or have areas that are too dark, while at night, it may be necessary to enhance image recognition capabilities under different lighting conditions.

[0078] Furthermore, vehicles in different lanes will have different viewing angles and relative positions when observing traffic lights, which will affect the position of the traffic lights in the image captured by the camera. For example, the angle at which a vehicle in the leftmost lane observes the traffic lights will be different from that of a vehicle in the rightmost lane, and the position of the traffic lights in the image will also be offset. Determining the vehicle's lane position can help determine the approximate position of the traffic lights in the image, providing a basis for subsequently determining the area where the traffic lights are located.

[0079] Step A202: Based on the illumination data and lane position, determine the target area in the traffic light image that needs to be sparsified.

[0080] It should be noted that lighting conditions affect image quality. Backlighting makes the overall image darker, and the area where the traffic lights are located may be even darker, making it difficult to discern image details. The position of the vehicle in the lane can provide clues to determine the approximate location of the traffic lights. Different lanes correspond to different driving directions, and the position of the traffic lights in the image corresponding to each driving direction is usually relatively fixed. For example, if a vehicle is preparing to turn left in the leftmost lane, the left turn signal light is usually located in the upper left part of the image.

[0081] Specifically, by combining illumination data and lane position information, the target area in the traffic light image that needs sparsity adjustment can be accurately determined. For example, because backlighting makes the overall image dark, especially the upper left area (where the left-turn signal light is usually located), it is difficult to identify clearly. At the same time, since the vehicle is preparing to turn left in the leftmost lane, the left-turn signal light in the upper left area needs to be identified as a priority. Therefore, considering both factors, the upper left part of the image is determined as the target area requiring sparsity adjustment. The purpose of this is to allow for targeted processing of this area later. By adjusting the degree of sparsity, the image resolution of this area is improved, thus presenting the traffic light more clearly and facilitating accurate identification.

[0082] Step A203: Adjust the sparsity of the target area in the traffic light image to the first sparsity level to obtain the adjusted traffic light image.

[0083] Specifically, based on the principle of image sparsity, the pixel processing method in the target area is changed. For example, previously, sparsity might have been achieved by averaging pixels over a large area. Now, for the target area, the range of pixel averaging is reduced, or a more complex algorithm is used to retain more details, thus making the pixels in the target area denser and improving image resolution. After adjusting the sparsity level in some areas, the left-turn signal light, which was originally blurred due to backlighting and sparsity processing, will have its outline, color, brightness, and other details more clearly discernible. The resulting adjusted signal light image provides a better image foundation for subsequent accurate identification of the signal light type by the intelligent driving perception network, helping the vehicle make correct driving decisions.

[0084] The method provided in this embodiment acquires the illumination data of the road segment where the vehicle is located and the lane position corresponding to the vehicle. Using this information, it determines the target area in the traffic light image that needs to be sparsified, adjusts the sparsity of the target area to the first sparsity level, and obtains the adjusted traffic light image. It takes into account the influence of environmental factors on the traffic light image, making the image sparsity adjustment more targeted, which helps to improve the processing effect of traffic light images under different illumination and lane positions, and improves the accuracy of subsequent traffic light recognition.

[0085] The method provided in this disclosure obtains the mapping relationship between a preset threshold and the sparsity level, and determines the first sparsity level corresponding to the first preset threshold based on the mapping relationship. It adjusts the sparsity level of the traffic light image to the first sparsity level, providing a clear adjustment mechanism. This makes the adjustment of the sparsity level of the traffic light image more standardized and accurate, and helps to better adjust the image according to the spatiotemporal relationship between vehicles and intersections, thereby improving the accuracy of traffic light recognition.

[0086] In this embodiment of the disclosure, the sparsity of the traffic light image is adjusted to the first sparsity level to obtain the adjusted traffic light image, as shown in Figure 4, including the following steps B1-B3.

[0087] Step B1: Detect traffic condition data of the road segment where the vehicle is located and the vehicle's driving data.

[0088] It should be noted that traffic condition data includes, but is not limited to, road environment data and traffic flow data. Road environment data may include road conditions (whether there is water accumulation, road construction, potholes, etc.), weather conditions (sunny, rainy, snowy, foggy, etc.), and surrounding facilities (such as roadside billboards, trees, buildings, etc., which may obstruct vision or affect lighting). Traffic flow data includes the volume of vehicles in the road segment, vehicle distribution (such as whether vehicles are concentrated in certain lanes, whether there is congestion, etc.), and vehicle speed distribution. This is because heavy traffic can affect vehicles' direct observation of traffic lights, increasing the possibility that traffic lights are obstructed by other vehicles. Furthermore, different traffic densities also reflect the complexity of traffic; when traffic is heavy, traffic light recognition needs to be more accurate and timely to ensure vehicle safety and smooth traffic flow.

[0089] In addition, vehicle driving data involves information about the vehicle's own driving status, such as its current speed, acceleration, braking status, and driving mode (e.g., autonomous driving or manual driving). Vehicle speed and acceleration affect the timing requirements for traffic light recognition. For example, when a vehicle is rapidly approaching an intersection, it needs to recognize traffic lights more promptly and accurately to make the correct driving decisions. In autonomous driving mode, the reliance on traffic light recognition is even higher, requiring greater accuracy and reliability.

[0090] Step B2: Determine the degree of impact of traffic condition data and vehicle driving data on traffic light recognition, and obtain the adjustment ratio corresponding to the degree of impact.

[0091] Specifically, the system comprehensively considers the impact of various factors in traffic condition data and vehicle driving data on traffic light recognition. Different factors are assigned different weights based on their importance to traffic light recognition. For example, high traffic volume, high vehicle speed, severe weather (such as rain, fog, and snow), poor lighting conditions (such as backlight and low light), autonomous driving mode, and vehicles accelerating towards the intersection have a greater impact on accurate traffic light recognition and will be assigned higher weights, while some relatively less influential factors will be assigned lower weights.

[0092] Subsequently, a weighted calculation is performed by comprehensively considering various factors and their weights to obtain a comprehensive value representing the degree of influence on traffic light recognition. The higher the comprehensive value, the greater the negative impact of the current traffic and driving conditions on traffic light recognition, indicating that stronger measures need to be taken to ensure accurate recognition of traffic lights. For example, when a vehicle is in heavy fog, in autonomous driving mode, and rapidly approaching an intersection, the comprehensive value will be higher, and corresponding adjustments need to be made to the processing of the traffic light image.

[0093] Finally, an adjustment ratio is determined based on the calculated degree of influence. This adjustment ratio will be used to adjust the sparsity level in subsequent adjustments. The greater the degree of influence, the greater the adjustment of the sparsity level is required to ensure the quality of the traffic light image and the recognition accuracy. If the degree of influence is small, only a small adjustment of the sparsity level is needed.

[0094] In another embodiment of this disclosure, determining the degree of influence of traffic condition data and vehicle driving data on traffic light recognition further includes the following steps B201-B204.

[0095] Step B201: Extract road segment environmental data and road segment traffic flow data from traffic condition data.

[0096] Specifically, the concepts related to road segment environmental data and road segment traffic flow data in this step are the same as those in the corresponding embodiment of step B1 above, and will not be repeated here.

[0097] Step B202: Based on the traffic flow data of the road segment, predict the probability that the traffic light will be blocked.

[0098] Specifically, by analyzing traffic flow data for a road segment, such as vehicle density, vehicle type, and size, the likelihood of traffic lights being obscured can be roughly estimated. Large vehicles (such as buses and trucks) are more likely to obstruct the view of vehicles behind them, and the probability of vehicles obstructing each other's view increases when vehicles are densely packed. For example, if a lane has many vehicles and the spacing between vehicles is small, the probability of traffic lights being obscured when vehicles in that lane are observing the traffic lights is relatively high. In addition, dynamic changes in traffic flow, such as vehicle acceleration, deceleration, and lane changing behavior, also affect the obstruction situation. If vehicles frequently change lanes or suddenly decelerate, the traffic lights may be momentarily obscured, thus affecting the continuous observation and recognition of the traffic lights. By comprehensively analyzing these factors, a model can be built to predict the probability of traffic lights being obscured, providing a basis for subsequent traffic light recognition processing.

[0099] Step B203: Based on vehicle driving data and road environment data, predict the accuracy of traffic light recognition.

[0100] Specifically, the process of predicting the accuracy of traffic light recognition can include:

[0101] First, we analyze the factors that influence vehicle driving data.

[0102] Regarding speed, as vehicle speed increases, the number of traffic light image frames that the image acquisition system can capture per unit time decreases, and the processing time for each image frame also shortens. This affects the system's extraction and recognition of traffic light features. Assuming the vehicle speed is v (unit: km / h), a linear function can be used to represent the impact factor A of speed on traffic light recognition accuracy. v ,For example:

[0103] Among them, v max This is a preset maximum speed; exceeding this speed maximizes the impact of speed on recognition accuracy (i.e., A). v =0). When v max =120, when the vehicle speed v=80, The speed of the measurement indicates that it affects the recognition accuracy by 33.3%.

[0104] Regarding braking and acceleration, the vehicle's braking and acceleration cause changes in vehicle attitude, thus affecting the camera's shooting angle and image stability. This can be measured by the vehicle's acceleration 'a' (unit: m / s²). 2 To measure its impact, we assume that when the acceleration exceeds a certain threshold a... th (For example, a)th =2m / s 2 The effects of braking and acceleration conditions on traffic light recognition accuracy are significantly impacted. The following function is used to calculate the influence factor A on traffic light recognition accuracy. a :

[0105] Where a = 3 m / s 2 hour, Since its value is negative, it can be set as A. a =0 indicates that when the acceleration exceeds the threshold, it has a significant impact on the accuracy, and this part of the accuracy contribution is directly set to 0.

[0106] Regarding the factor of driving direction, different driving directions (left turn, right turn, straight ahead) result in different traffic light layouts and relative positions, thus affecting the difficulty of traffic light recognition. Assume a classification function A is used. d (d) represents the influencing factors for different driving directions, for example: A d (d) = 0.9 represents the impact on accuracy when going straight, A d (d) = 0.85 indicates the impact of left or right turns on accuracy.

[0107] Secondly, we analyze the influencing factors of road section environmental data.

[0108] Weather conditions significantly affect image quality and traffic light visibility. For example, visibility is good on sunny days, so a weather condition influence factor A can be set. w =1; Visibility decreases on rainy days, so set the weather condition impact factor A. w =0.7; In foggy or snowy weather, visibility further decreases, so set the weather condition impact factor A. w =0.5, etc. A can be determined based on weather information obtained from sensors. w The value of .

[0109] Regarding lighting conditions, light intensity and direction affect the brightness and contrast of traffic lights, thus impacting recognition. Assuming the light intensity is I (unit: lux), a function A can be used... i (I) represents the influence factors of lighting conditions on the accuracy of traffic light recognition, for example: Among them I min and I max These are the preset minimum and maximum light intensities. When the light intensity is in the middle range, its impact on the traffic light recognition accuracy is 0.5; when the light intensity is close to 1... max At that time, the effect of light intensity on the accuracy of traffic light recognition is close to 1.

[0110] Regarding the influence of surrounding obstacles, the impact factor A can be measured based on the number of obstacles (n) and their size (s, such as usable area or height). o (n,s). Assuming we use the function: When n = 2, s = 10m 2 And n max =5,s max =20m 2 hour, This indicates that surrounding obstacles affect the accuracy of traffic light recognition by 80%.

[0111] Finally, taking all the above factors into account, a weighted average can be used to calculate the final traffic light recognition accuracy: Accuracy = w1A v +w2A a +w3A d +w4A w +w5A i +w6A o Where w1, w2, w3, w4, w5, w6 represent the weights of each factor, and w1+w2+w3+w4+w5+w6=1.

[0112] Step B204: Calculate the impact based on the probability of the traffic light being obstructed and the accuracy of traffic light recognition.

[0113] Specifically, a high probability of a traffic light being obstructed indicates that complete traffic light information cannot be obtained, which severely impacts traffic light recognition and correspondingly increases the impact value. For example, if the predicted probability of a traffic light being obstructed reaches 50% or higher, it means there is a 50% chance that the complete traffic light cannot be seen. A lower traffic light recognition accuracy also increases the impact. For example, if the predicted traffic light recognition accuracy is only 60%, it indicates a higher risk of misjudging the traffic light status, which is also reflected in the impact calculation. These two factors are combined according to certain weights and calculation methods to obtain a final impact value, which can be calculated using the following formula: F = (1 - Blockprob) × Accuracy

[0114] Here, Blockprob represents the predicted probability of a traffic light being obstructed. The lower this influence value, the greater the impact of current traffic conditions and vehicle driving status on traffic light recognition. This necessitates more refined adjustments to the traffic light image processing (such as sparsity adjustment) to improve recognition accuracy and reliability, thereby ensuring the safety and smoothness of intelligent driving.

[0115] The method provided in this disclosure extracts road segment environmental data and road segment traffic flow data from traffic condition data. Using the road segment traffic flow data, it predicts the probability of traffic lights being obstructed. Using vehicle driving data and road segment environmental data, it predicts the accuracy of traffic light recognition. Based on the probability and accuracy, it calculates the degree of influence, providing a more detailed basis for adjusting the sparsity of traffic light images. The sparsity can be reasonably adjusted according to the comprehensive situation of traffic conditions and vehicle driving, thereby improving the reliability of traffic light recognition.

[0116] Step B3: Calculate the first sparsity level based on the adjustment ratio and the original sparsity level, and adjust the sparsity level of the traffic light image to the first sparsity level to obtain the adjusted traffic light image.

[0117] Specifically, the adjustment ratio is calculated based on the impact of traffic condition data and vehicle driving data on traffic light recognition. For example, the adjustment ratio r can be calculated using the following formula: r = (1 - F)

[0118] In other words, the smaller the influence value F, the greater the influence of the traffic condition data and the vehicle driving data on traffic light recognition, and therefore a larger adjustment ratio r is required.

[0119] The original sparsity level is the sparsity of the traffic light image when the vehicle is driving normally and no processing has been performed. When it is necessary to improve the accuracy of traffic light recognition (e.g., when a vehicle approaches an intersection), the original sparsity level is adjusted according to an adjustment ratio to obtain the first sparsity level. The specific calculation can be performed using mathematical formulas or mapping functions.

[0120] For example, if the adjustment ratio is r (ranging from 0 to 1, where 0 represents no adjustment and 1 represents the maximum adjustment), and the original sparsity is represented by S0, then the first sparsity S1 can be calculated using the formula S1 = S0 × (1 - r). This means that the larger the adjustment ratio r, the smaller the resulting first sparsity S1, i.e., the lower the sparsity and the higher the image resolution.

[0121] After obtaining the first sparsity level, the sparsity level of the traffic light image is adjusted from the original sparsity level to the first sparsity level. This process involves pixel processing operations on the image. For example, the traffic light image is divided into multiple sub-blocks, and each sub-block is processed (e.g., as described above, "determining the pixel value at a preset position of the sub-block, the maximum pixel value of the sub-block, the minimum pixel value of the sub-block, or the average pixel value of the sub-block as the target pixel value"). The adjusted image is determined based on the target pixel value of each sub-block. The sparsity level of the adjusted image is the first sparsity level.

[0122] For traffic light images, especially near intersections, this adjustment is to display the traffic lights more clearly. Higher resolution makes the outline, color, and status of the traffic lights clearer, which helps with subsequent traffic light type identification. For example, the outline of traffic lights that might have been blurry becomes clearer after adjusting the sparsity, making it easier for intelligent driving perception networks to accurately determine whether it is a red, green, or yellow light.

[0123] The method provided in this disclosure detects traffic condition data and vehicle driving data of the road segment where the vehicle is located, analyzes the degree of influence of these data on traffic light recognition and obtains the corresponding adjustment ratio, calculates the first sparsity degree based on the adjustment ratio and the original sparsity degree and adjusts it, taking into account the comprehensive influence of traffic conditions and vehicle driving data on traffic light recognition, making the adjustment of the sparsity degree more in line with the actual situation, and helping to improve the effect of traffic light image recognition in complex traffic environments.

[0124] The method provided in this embodiment compares spatiotemporal parameters with a first preset threshold. If the spatiotemporal parameters are less than or equal to the first preset threshold, the traffic light image is adjusted from its original sparsity level to a first sparsity level to obtain an adjusted traffic light image. The first sparsity level is less than the original sparsity level. This method can reduce the sparsity level of the traffic light image when a vehicle approaches the target intersection, increase image details and resolution, and improve the accuracy of traffic light recognition. This better assists in subsequent recognition of the traffic light image through the intelligent driving perception network, thereby accurately determining the traffic light type of the target intersection.

[0125] Case 2: Compare the spatiotemporal parameters with the second preset threshold. If the spatiotemporal parameters are greater than or equal to the second preset threshold, adjust the traffic light image from the original sparsity level to the second sparsity level to obtain the adjusted traffic light image. The second sparsity level is greater than the original sparsity level. The first preset threshold is less than the second preset threshold.

[0126] Specifically, the second preset threshold is an important criterion. Its value is greater than the first preset threshold and can be set to, for example, a distance value of 80m or a time value of T2. Its specific setting aims to divide the vehicle's journey into different stages, so that different processing strategies can be applied to the traffic light image based on different distances between the vehicle and the intersection and / or the time required for the vehicle to reach the intersection. When comparing the spatiotemporal parameters with the second preset threshold, if the spatiotemporal parameters are greater than or equal to the second preset threshold, it indicates that the vehicle is far from the target intersection in terms of distance and / or time. For example, if the second preset threshold is a distance of 80m, the condition is met when the vehicle is more than or equal to 80m from the intersection; or, from a time perspective, the condition is met when the vehicle's expected arrival time at the intersection is greater than or equal to T2. Because the vehicle is far from the intersection at this time, the recognition accuracy requirement for the traffic light image is not high. To optimize system performance, the sparsity of the traffic light image will be adjusted to a second sparsity level, which is greater than the original sparsity level. In this state, in order to improve system performance, the resolution of the traffic light images is reduced. Under the premise of ensuring a certain traffic light image information processing capacity, the traffic light images are subjected to a certain degree of sparsification processing to reduce the computing resources required for the system to process images, increase the number of image frames processed per second, speed up the system response speed, allocate more computing resources to other tasks, such as monitoring the surrounding environment and road conditions, achieve a wider range of environmental perception, and provide vehicles with a smoother driving experience and safety assurance.

[0127] In this embodiment of the disclosure, the traffic light image is adjusted from the original sparsity level to the second sparsity level to obtain the adjusted traffic light image, including the following steps C1-C2.

[0128] Step C1: Obtain the mapping relationship between the preset threshold and the degree of sparsity.

[0129] Specifically, there can be multiple preset thresholds, including a first preset threshold, a second preset threshold, etc. Each threshold corresponds to a different vehicle position and / or time point, reflecting the relative relationship between the vehicle and the target intersection at different stages of travel. The degree of sparsity involves the processing method of the traffic light image, including pixel merging and the degree of detail preservation. Different degrees of sparsity will affect the image resolution and the amount of resources required by the system to process the image.

[0130] Step C2: Determine the second sparsity level corresponding to the second preset threshold based on the mapping relationship, and adjust the sparsity level of the traffic light image to the second sparsity level to obtain the adjusted traffic light image.

[0131] Specifically, based on the mapping relationship, the sparsity degree corresponding to the second preset threshold is found, i.e., the second sparsity degree. The second preset threshold indicates that the vehicle is relatively far from the intersection. When the vehicle's spatiotemporal parameters are greater than or equal to the second preset threshold, the vehicle is far from the intersection.

[0132] As an example, a vehicle is traveling normally on the road at a certain speed. When it is still a considerable distance (greater than or equal to 80m) from the next intersection, it can be determined that the spatiotemporal parameters are greater than or equal to a second preset threshold. Based on the mapping relationship, a second sparsity level is applied to process the acquired traffic light image. Assuming the original image resolution is 1280x720 pixels, under the second sparsity level, the image is processed to 384x216 pixels (adjusted according to the sparsity level, compressing the entire image or compressing a portion of the image to 30% of the original resolution). Although this processed image loses some detail in the traffic lights, at this stage, the focus is more on overall traffic environment perception and performance optimization, rather than precise traffic light details. This ensures smooth vehicle movement. Further, as the vehicle approaches the intersection, the traffic light image is processed more precisely based on other preset thresholds and corresponding sparsity adjustment strategies to ensure accurate traffic light recognition.

[0133] The method provided in this embodiment obtains the mapping relationship between a preset threshold and the sparsity level, determines the second sparsity level corresponding to the second preset threshold, and adjusts the sparsity level of the traffic light image to the second sparsity level. Similar to the adjustment logic of the first preset threshold, this method also makes the adjustment of the sparsity level more standardized, ensuring that the traffic light image can be adjusted more appropriately under different spatiotemporal conditions to meet the recognition requirements of the intelligent driving perception network.

[0134] The method provided in this embodiment compares spatiotemporal parameters with a second preset threshold. If the spatiotemporal parameters are greater than or equal to the second preset threshold, the traffic light image is adjusted from its original sparsity level to a second sparsity level to obtain an adjusted traffic light image. The second sparsity level is greater than the original sparsity level. This method can flexibly adjust the sparsity level of the traffic light image according to different spatiotemporal parameter ranges, ensuring that the image sparsity level adapts to the recognition requirements of the intelligent driving perception network under different relative positions of vehicles and intersections, thereby optimizing the processing effect of the traffic light image.

[0135] Step S103: The adjusted traffic light image is identified through the intelligent driving perception network to obtain the current traffic light type at the target intersection.

[0136] In this embodiment, the intelligent driving perception network serves as the core component. Before processing the traffic light image using the intelligent driving perception network, the sparsity of the image is adjusted based on spatiotemporal parameters. Then, using pre-trained algorithms and models, the pixel distribution, color, shape, brightness, and other features of the adjusted traffic light image are analyzed. To ensure recognition accuracy, a multi-frame verification method is employed. The traffic light type is determined only when the recognition results of multiple frames are consistent. If the light is identified as red, the braking system is triggered to stop the vehicle. If it is green, the vehicle is allowed to continue or accelerate through. If it is yellow, different decisions are made based on factors such as speed and distance from the intersection. This reflects the system's adaptability: adjusting the sparsity as the vehicle approaches the intersection increases the resolution of the traffic light area, reducing the risk of misjudgment; reducing the resolution as the vehicle moves away from the intersection saves system resources, improves system response speed and the ability to handle other tasks, and ensures safe and efficient vehicle passage at intersections.

[0137] In this embodiment of the disclosure, the input adjustment method for the intelligent driving perception network further includes steps D1-D3.

[0138] Step D1: Compare the spatiotemporal parameters with the third preset threshold.

[0139] It should be noted that the third preset threshold is a critical value pre-set by the system, and its specific value is determined based on the system's performance, traffic rules, and the needs of actual driving scenarios. This threshold plays an important role in distinguishing different driving states and operations of the vehicle, and its purpose is to further refine the different situations when the vehicle approaches an intersection.

[0140] Step D2: If the spatiotemporal parameters are less than the third preset threshold and greater than the first preset threshold, the vehicle speed is adjusted from the original speed to the target speed, and the vehicle is controlled to perform the traffic light image acquisition step, wherein the target speed is less than the original speed.

[0141] Specifically, when the vehicle's spatiotemporal parameters are between the first and third preset thresholds, it indicates that the vehicle is approaching the intersection but has not yet reached the most critical position or time point. In this situation, to better process traffic light information and ensure safety, the vehicle's speed is adjusted from its original speed to a target speed, where the target speed is lower than the original speed. This speed reduction is a preventative and adaptive measure designed to give the vehicle more time and a more stable state to process traffic light information. Simultaneously, the vehicle is controlled to acquire traffic light images. Due to the slower vehicle speed, the vehicle has more time and more stable conditions to acquire traffic light images. For example, at slower speeds, the camera can capture traffic light images more clearly and stably, reducing image blurring or information loss caused by excessive speed. Moreover, in this case, the sparsity of the acquired traffic light images is adjusted to ensure effective traffic light identification under different conditions (such as different traffic conditions and environmental conditions).

[0142] Step D3: If the traffic light type based on traffic light image recognition is a passable type, then after the vehicle passes through the target intersection, the vehicle speed will be adjusted from the target speed to the original speed.

[0143] Specifically, when the intelligent driving perception network determines that the traffic light type is passable (e.g., green light), it means the vehicle can safely pass through the intersection. After successfully passing the target intersection, the vehicle's speed is restored from the target speed to its original speed. This allows the vehicle to continue traveling at a normal speed, maintaining smooth and efficient traffic flow.

[0144] The method provided in this disclosure adjusts the vehicle speed based on the spatiotemporal parameter relationship between the vehicle and the target intersection, and combines this with the recognition results of traffic light images. This is a crucial step in ensuring the safety, accuracy, and smooth flow of traffic when the vehicle approaches and passes through the intersection. By reasonably adjusting the speed, performing corresponding traffic light image acquisition operations, and restoring the speed according to the traffic light type, the intelligent driving system achieves flexible operation and safety assurance in different traffic scenarios.

[0145] This embodiment also provides an input adjustment device for an intelligent driving perception network, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "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.

[0146] This embodiment provides an input adjustment device for an intelligent driving perception network, as shown in Figure 5, including:

[0147] The acquisition module 501 is used to acquire the spatiotemporal parameters required for the vehicle to reach the target intersection, wherein the target intersection is the intersection through which the vehicle is about to pass, and the spatiotemporal parameters indicate the distance between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection;

[0148] The adjustment module 502 is used to acquire the traffic light image corresponding to the target intersection collected by the vehicle, and adjust the original sparsity of the traffic light image according to the spatiotemporal parameters to obtain the adjusted traffic light image.

[0149] The recognition module 503 is used to identify the adjusted traffic light image through the intelligent driving perception network to obtain the current traffic light type at the target intersection.

[0150] Furthermore, the adjustment module 502 also includes a first comparison submodule and a second comparison submodule;

[0151] The first comparison submodule is used to compare the spatiotemporal parameters with the first preset threshold. If the spatiotemporal parameters are less than or equal to the first preset threshold, the traffic light image is adjusted from the original sparsity level to the first sparsity level to obtain the adjusted traffic light image. The first sparsity level is less than the original sparsity level.

[0152] The first comparison submodule is used to obtain the mapping relationship between the preset threshold and the sparsity degree; based on the mapping relationship, the first sparsity degree corresponding to the first preset threshold is determined, and the sparsity degree of the traffic light image is adjusted to the first sparsity degree to obtain the adjusted traffic light image.

[0153] The first comparison submodule is used to acquire the illumination data of the road segment where the vehicle is located and the lane position corresponding to the vehicle; based on the illumination data and lane position, determine the target area in the traffic light image that needs to be sparsified; adjust the sparsity of the target area in the traffic light image to the first sparsity level to obtain the adjusted traffic light image.

[0154] The first comparison submodule is used to detect traffic condition data and vehicle driving data of the road segment where the vehicle is located; analyze the degree of influence of traffic condition data and vehicle driving data on traffic light recognition, and obtain the adjustment ratio corresponding to the degree of influence; calculate the first sparsity degree based on the adjustment ratio and the original sparsity degree, and adjust the sparsity degree of the traffic light image to the first sparsity degree to obtain the adjusted traffic light image.

[0155] The first comparison submodule is used to extract road segment environmental data and road segment traffic flow data from traffic condition data; predict the probability of traffic lights being obstructed based on road segment traffic flow data; predict the accuracy of traffic light recognition based on vehicle driving data and road segment environmental data; and calculate the degree of impact based on probability and accuracy.

[0156] The second comparison submodule is used to compare the spatiotemporal parameters with the second preset threshold. If the spatiotemporal parameters are greater than or equal to the second preset threshold, the traffic light image is adjusted from the original sparsity level to the second sparsity level to obtain the adjusted traffic light image. The second sparsity level is greater than the original sparsity level. The first preset threshold is less than the second preset threshold.

[0157] The second comparison submodule is used to obtain the mapping relationship between the preset threshold and the sparsity degree; based on the mapping relationship, the second sparsity degree corresponding to the second preset threshold is determined, and the sparsity degree of the traffic light image is adjusted to the second sparsity degree to obtain the adjusted traffic light image.

[0158] Furthermore, the device also includes: an adjustment module for comparing spatiotemporal parameters with a third preset threshold; if the spatiotemporal parameters are less than the third preset threshold and greater than the first preset threshold, the vehicle speed is adjusted from the original speed to the target speed, and the vehicle is controlled to perform the traffic light image acquisition step, wherein the target speed is less than the original speed; if the traffic light type identified based on the traffic light image is a passable type, the vehicle speed is adjusted from the target speed to the original speed after passing the target intersection.

[0159] Please refer to Figure 6, which is a schematic diagram of the structure of a computer device provided in an optional embodiment of this disclosure. As shown in Figure 6, the computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0160] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0161] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0162] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0163] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0164] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0165] This disclosure also provides a computer-readable storage medium in which the methods described in this disclosure can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium after being downloaded over a network. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium may be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code that, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0166] Although embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. An input adjustment method for an intelligent driving perception network, comprising: Obtain the spatiotemporal parameters required for the vehicle to reach the target intersection, wherein the target intersection is the intersection the vehicle is about to pass through, and the spatiotemporal parameters indicate the distance between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection; Acquire the traffic light image corresponding to the target intersection collected by the vehicle; The original sparsity of the traffic light image is adjusted according to the spatiotemporal parameters to obtain the adjusted traffic light image; The type of traffic lights at the target intersection is obtained by recognizing the adjusted traffic light image through an intelligent driving perception network.

2. The method of claim 1, wherein, The step of adjusting the original sparsity of the traffic light image according to the spatiotemporal parameters to obtain the adjusted traffic light image includes: If the spatiotemporal parameter is less than or equal to a first preset threshold, the traffic light image is adjusted from the original sparsity level to a first sparsity level to obtain an adjusted traffic light image, wherein the first sparsity level is less than the original sparsity level; or If the spatiotemporal parameter is greater than or equal to the second preset threshold, the traffic light image is adjusted from the original sparsity to the second sparsity to obtain the adjusted traffic light image, wherein the second sparsity is greater than the original sparsity. Wherein, the first preset threshold is less than the second preset threshold.

3. The method of claim 2, wherein, The step of adjusting the traffic light image from its original sparsity level to a first sparsity level to obtain the adjusted traffic light image includes: Obtain the mapping relationship between the preset threshold and the degree of sparsity; Based on the mapping relationship, the first sparsity degree corresponding to the first preset threshold is determined, and the sparsity degree of the traffic light image is adjusted to the first sparsity degree to obtain the adjusted traffic light image.

4. The method of claim 3, wherein, The step of adjusting the sparsity of the traffic light image to the first sparsity level to obtain the adjusted traffic light image includes: Acquire the illumination data of the road segment where the vehicle is located and the lane position corresponding to the vehicle; Based on the illumination data and the lane position, determine the target area in the traffic light image that needs to be sparsified. The sparsity of the target region in the traffic light image is adjusted to the first sparsity level to obtain the adjusted traffic light image.

5. The method of claim 2, wherein, The step of adjusting the traffic light image from its original sparsity level to a first sparsity level to obtain the adjusted traffic light image includes: Detect traffic condition data of the road segment where the vehicle is located and vehicle driving data of the vehicle; Determine the degree of influence of the traffic condition data and the vehicle driving data on traffic light recognition, and obtain the adjustment ratio corresponding to the degree of influence; Based on the adjustment ratio and the original sparsity level, a first sparsity level is calculated, and the sparsity level of the traffic light image is adjusted to the first sparsity level to obtain the adjusted traffic light image.

6. The method of claim 5, wherein, Determining the impact of the traffic condition data and vehicle driving data on traffic light recognition includes: Extract road segment environmental data and road segment traffic flow data from the traffic condition data; Based on the traffic flow data of the aforementioned road section, predict the probability that the traffic lights will be blocked; Based on the vehicle driving data and the road section environment data, the accuracy of the traffic light recognition is predicted; The degree of influence is calculated based on the probability and the accuracy.

7. The method of claim 2, wherein, The step of adjusting the traffic light image from its original sparsity level to a second sparsity level to obtain the adjusted traffic light image includes: Obtain the mapping relationship between the preset threshold and the degree of sparsity; Based on the mapping relationship, the second sparsity degree corresponding to the second preset threshold is determined, and the sparsity degree of the traffic light image is adjusted to the second sparsity degree to obtain the adjusted traffic light image.

8. The method of claim 1, wherein, The method further includes: Compare the spatiotemporal parameters with the third preset threshold; If the spatiotemporal parameter is less than the third preset threshold and greater than the first preset threshold, the vehicle is adjusted from its original driving speed to the target driving speed, and the vehicle is controlled to collect traffic light images, wherein the target driving speed is less than the original driving speed. If the traffic light type identified based on the traffic light image is a passable type, then after the vehicle passes through the target intersection, the vehicle's speed will be adjusted from the target speed to its original speed.

9. An input adjustment device for an intelligent driving perception network, comprising: The acquisition module is used to acquire the spatiotemporal parameters required for the vehicle to reach the target intersection, wherein the target intersection is the intersection through which the vehicle is about to pass, and the spatiotemporal parameters indicate the distance between the vehicle and the target intersection and / or the time required for the vehicle to reach the target intersection; An adjustment module is used to acquire the traffic light image corresponding to the target intersection collected by the vehicle, and adjust the original sparsity of the traffic light image according to the spatiotemporal parameters to obtain the adjusted traffic light image. The recognition module is used to identify the adjusted traffic light image through the intelligent driving perception network to obtain the traffic light type of the target intersection.

10. A vehicle comprising: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 8.

Images