A computer vision-based vehicle body height adjustment method

Abstract

The application discloses a vehicle body height adjusting method based on computer vision, establishes a traffic signal lamp recognition model and pre-training; continuously obtains vehicle speed information from a bus and compares with critical speed; when the vehicle speed reaches a judgment condition, a corresponding counter is added by one, and when the counter reaches a set value, visual detection is executed; when the vehicle speed is on a critical speed or between two critical speeds, the counter counts, and when the counter reaches the set value, the three counters in this mode are all cleared; the visual detection is to detect the environment in front of the vehicle; the obtained data is loaded to the trained traffic signal lamp recognition model for detection, and whether height adjustment is executed is determined according to the detection result; when no traffic signal lamp is detected, height adjustment is performed; and when a traffic signal lamp is detected, the height is kept without adjustment. The application has the advantage that frequent and useless height adjustment caused by fluctuation on the critical speed is avoided.

Description

Technical Field

[0001] This invention relates to the field of electronically controlled air suspension systems, and in particular to a method for adjusting vehicle height based on computer vision. Background Technology

[0002] Commercial vehicles are a vital component of road transportation, and improving their performance is of paramount importance. One crucial parameter for commercial vehicles is vehicle height. During operation, it's often difficult to simultaneously meet the demands of high-speed stability, fuel efficiency, and maneuverability in complex road conditions. Therefore, electronically controlled air suspension (ECAS) systems have been developed to achieve variable vehicle height. A key component of ECAS is the vehicle height adjustment method. Traditional methods link vehicle height to speed. When the vehicle starts, it enters an economy mode. In this mode, vehicle height adjustment is linked to a specific speed; above this speed, the vehicle height lowers to the economy level; below this speed, it rises to the default height. In poor road conditions, the driver actively determines whether to enter an obstacle-crossing mode. In this mode, the system uses a specific speed as a benchmark; below this speed, the vehicle height increases to the obstacle-crossing height; above this speed, it adjusts to the default height. However, when the vehicle speed is near the critical speed for switching vehicle height, speed fluctuations cause the system to frequently adjust the vehicle height, resulting in unnecessary wear and tear and risks to the system and its components. Therefore, finding a superior vehicle height adjustment method is extremely important. However, even with improved vehicle height adjustment methods, commercial vehicles still experience frequent height adjustments when their speed fluctuates significantly within a short period. This is especially true at intersections with traffic lights, where vehicles wait for the light to turn red. During this process, the vehicle's speed drops from its normal speed to zero, waits briefly, and then accelerates back to normal speed. This results in at least two height adjustments within 1-2 minutes, a situation that should be avoided as much as possible from both a driving experience and energy conservation perspective. To address this issue, this paper proposes a vehicle height adjustment method that integrates traffic light information using computer vision technology. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a computer vision-based vehicle height adjustment method to balance the stability and economy at high speeds and the vehicle passability in complex road conditions during commercial vehicle operation, while avoiding frequent vehicle height adjustments at intersections with traffic lights.

[0004] This invention provides a computer vision-based method for adjusting vehicle body height, comprising the following steps: S1: Establish a YOLOv4-tiny traffic light recognition model, perform pre-training and secondary pre-training, collect, process, label, and divide data, and use the self-collected dataset for training; S2: The electronically controlled air suspension system (ECAS) continuously sends vehicle speed detection commands, obtains vehicle speed information from the CAN bus every set time interval Δt, and compares it with the critical speed. When the vehicle speed reaches the judgment condition, the corresponding counter increments by one, while the counters corresponding to other execution states are cleared to zero. When the counter reaches the set value, step S3 visual detection is executed. When the vehicle speed is at or between the two critical speeds, the counter counts. When the counter reaches the set value, all three counters in this mode are cleared to zero. S3: Visual detection detects the environment in front of the vehicle; the data acquired by the camera is loaded into the trained YOLOv4-tiny traffic light recognition model for detection, and the height adjustment is determined based on the detection results. S4: If no traffic light is detected, adjust the height or switch modes; or if a traffic light is detected, keep the height unchanged and continue speed detection in this mode or switch to the default mode.

[0005] As a preferred method, visual inspection detects the environment in front of the vehicle by calling on the vehicle's onboard camera. The Basler acA2440-35uc camera is used to detect traffic lights, which has a high frame rate and resolution.

[0006] As a preferred option, the YOLOv4-tiny pre-trained model, which has excellent transferability, is used.

[0007] Ideally, the two pre-training sessions for the model should use the large public dataset ImageNet and the public traffic light dataset WPI.

[0008] As a preferred approach, a self-collected dataset should be established; the collected data should include images of traffic lights from multiple scenes, angles, and distances to improve the overall capabilities of the model.

[0009] As a preferred approach, the collected data is labeled, and then divided into training and validation sets in a 7:3 ratio.

[0010] The superior advantages of this invention are as follows: compared with the general commercial vehicle body height adjustment method, it not only simultaneously satisfies the stability and energy saving at high speeds and the passability under complex road conditions, but also avoids speed fluctuations at critical speeds and frequent body height adjustments at traffic light intersections, thereby enhancing the stability and efficiency of the system. Attached Figure Description

[0011] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of an embodiment of the present invention; Figure 2 This is a training hierarchy diagram of the YOLOv4-tiny traffic light recognition model according to an embodiment of the present invention. Implementation

[0012] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0013] Figure 1 A flowchart of an embodiment of the present invention is shown. Figure 2 The diagram illustrates the training hierarchy of the YOLOv4-tiny traffic light recognition model according to an embodiment of the present invention. Figure 1 As shown, the present invention provides a computer vision-based method for adjusting vehicle body height, comprising the following steps: Step S1: Establish a YOLOv4-tiny traffic light recognition model, including pre-training and secondary pre-training, self-collection, processing, labeling, and partitioning of data, and training using the self-collected dataset; the specific steps are as follows: Step S11, as follows Figure 2 As shown, a pre-trained model with good transfer performance published by the model developer is used. The model is pre-trained once using the large public dataset ImageNet and then pre-trained a second time using the public traffic light dataset WPI. Step S12: Collect images of traffic lights with multiple environments, angles, distances, and interferences. Perform preprocessing including filtering, deduplication, and noise reduction. Then, manually label the images and divide them into training and validation sets in a 7:3 ratio. Step S13: Train using the self-collected dataset.

[0014] Step S2: The system continuously issues vehicle speed detection commands, acquires vehicle speed information from the CAN bus every set time interval Δt, and compares it with the critical speed. When the vehicle speed reaches the judgment condition, the corresponding counter is incremented by one, while the counters corresponding to other execution states are cleared to zero. When the counter reaches the set value, step S3 visual detection is executed. When the vehicle speed is at the critical speed (i.e., default mode) or between two critical speeds (i.e., pass mode), the counter counts. When the counter reaches the set value, all three counters in this mode are cleared to zero. The specific steps are as follows: In step S21, the electronically controlled air suspension system ECAS continuously sends out vehicle speed detection commands and obtains vehicle speed information from the CAN bus every set time interval Δt. Step S22, (a) In the default mode, compare the vehicle speed with the critical speed v1: If the vehicle speed is greater than the critical speed v1, the corresponding operation is to adjust the vehicle height to the energy-saving height, increment the energy-saving height counter h2, and set the default height counter h1 and the critical speed counter v1 to zero. If the vehicle speed is less than the critical speed v1, the corresponding operation is to adjust the vehicle height to the default height, increment the default height counter h1, and set the energy-saving height counter h2 and the critical speed counter v1 to zero. If the vehicle speed is equal to the critical speed v1, then the critical speed counter v1 is incremented by one. When the aforementioned counter h2 or h1 reaches the set value, counter h2 or h1 is cleared to zero, and the visual inspection in step S3 is initiated; when counter v1 reaches the set value, all three counters h2, h1, and v1 in the default mode are set to zero. Critical speed v1 is the dividing line between the vehicle body operating at the default height and the energy-saving height; The set value reached by counter h2 or h1 is related to the desired speed range holding time. The product of the set time Δt and the set value of each counter is the desired speed range holding time under each condition. When the counter reaches the set value, the corresponding speed range has been held for the desired time, indicating that the corresponding operation needs to be performed. For example, if research shows that around 60 km / h is the dividing line between the vehicle's desired operating height and the energy-saving height, then v1 is set to 60 km / h, and it is believed that the speed holding time should be more than 30 seconds to clearly indicate that a change in height is necessary. Taking the speed extraction time interval Δt as 0.2 seconds, the set value of the counter is 30 ÷ 0.2 = 150. Assuming the vehicle is operating at its energy-saving altitude at 70 km / h, the driver suddenly realizes an emergency and needs to brake, reducing the speed to below 60 km / h. However, the driver quickly realizes this was a misjudgment and accelerates back to above 60 km / h. With a traditional strategy, this would require two altitude changes: one crossing the 60 km / h threshold, once from 70 to below 60, and once from below 60 to above 60. However, because the driver reacts quickly (the process takes only 25 seconds), the time below 60 km / h is less than the set 30-second holding time. The default altitude counter h1 starts counting after entering the 60 km / h range, but once the speed exceeds 60 km / h, the energy-saving altitude counter h2 starts counting, resetting h1 to zero and preventing further altitude changes, thus saving energy and durability. The counter v1 is used to detect speed fluctuations. Counter v1 does not have a corresponding height adjustment; it only detects fluctuations. This is equivalent to assuming that the vehicle speed fluctuates greatly during this period. Counters h1 and h2 cannot accurately reflect the vehicle speed, so all counters are refreshed. (ii) In the passage mode, compare the vehicle speed with the critical speeds v3 and v4: If the vehicle speed is greater than v4, the corresponding operation is to exit the pass mode and switch to the default mode, increment the exit counter v4 by one, and reset the pass height counter h3 and the zero counter v34 to zero. If the vehicle speed is less than v3, the corresponding operation is to adjust the vehicle height to the passing height, increment the passing height counter h3 by one, and exit counter v4 and zero counter v34. If the vehicle speed is between the critical speeds v3 and v4, the counter v34 is incremented by one; Where: V3 is the dividing line between the vehicle's pass-through height and other heights, and the vehicle is driving at this speed to indicate that it is crossing an obstacle; V4 is the threshold for entering and exiting the pass-through mode; and: v1>v4>v3; When counter v4 or h3 reaches the set value, the counter is cleared to zero and visual inspection begins; when counter v34 reaches the set value, all three counters in the pass mode are cleared to zero. The product of the set Δt time and the set value of each counter is the desired speed range holding time under each situation. When the counter reaches the set value, the corresponding vehicle speed range is held for the desired time, which means that the corresponding operation is indeed required. Figure 1 The "vehicle speed exceeding or falling below a certain speed and maintaining that speed for a certain period of time" mechanism is triggered by a corresponding counter reaching a set value. As mentioned earlier, the duration of the desired speed range is determined by Δt and the counter's set value. When the counter reaches the set value, it indicates that the vehicle speed range has been maintained for the desired time, and subsequent operations can be performed. If the counter does not reach the set value, subsequent visual detection will not be triggered, and the system will revert to the default mode. Step 23: Mode switching requires manual selection. Choose to switch from the default mode to the pass mode, or from the default mode to the height hold mode. When switching from the default mode to the pass mode, vehicle speed information is obtained from the CAN bus. If the vehicle speed is less than v4, the mode is switched to pass mode; if the vehicle speed is too high, the switch fails and returns to the default mode. When manually switching from the default mode to the height hold mode, the switch is performed directly. In the height hold mode, if the vehicle speed is greater than v1, the mode automatically returns to the default mode.

[0015] Step 3, visual detection, involves using the vehicle's onboard camera to detect the environment in front of the vehicle. The data acquired by the camera is then loaded into the trained YOLOv4-tiny traffic light recognition model for detection. Based on the detection results, a decision is made on whether to perform height adjustment. The following are two scenarios: 1) When a traffic light is detected ahead, maintain altitude and do not perform any altitude switching, continue vehicle speed detection in this mode; however, an exit mechanism is established: in the default mode, if the vehicle speed is greater than the critical speed v1, return to the default mode to continue detection; in the pass mode, if the vehicle speed is greater than v4, return to the default mode. 2) If no traffic light is detected, adjust the height or switch modes accordingly based on the vehicle speed information.

[0016] This invention uses computer vision to detect and identify traffic lights at intersections, maintaining vehicle height and avoiding unnecessary adjustments. It also optimizes the vehicle height adjustment method by maintaining the vehicle speed range for a certain period of time to determine whether to perform height adjustment, thus avoiding frequent and unnecessary height adjustments caused by speed fluctuations at critical speeds.

[0017] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A computer vision-based method for adjusting vehicle body height, comprising the following steps: S1: Establish a YOLOv4-tiny traffic light recognition model, perform pre-training and secondary pre-training, collect, process, label, and divide data, and use the self-collected dataset for training; S2: The electronically controlled air suspension system (ECAS) continuously sends vehicle speed detection commands, obtains vehicle speed information from the CAN bus every set Δt time, and compares it with the critical speed. Step S22, (a) In the default mode, compare the vehicle speed with the critical speed v1: If the vehicle speed is greater than the critical speed v1, the corresponding operation is to adjust the vehicle height to the energy-saving height, increment the energy-saving height counter h2, and set the default height counter h1 and the critical speed counter v1 to zero. If the vehicle speed is less than the critical speed v1, the corresponding operation is to adjust the vehicle height to the default height, increment the default height counter h1, and set the energy-saving height counter h2 and the critical speed counter v1 to zero. If the vehicle speed is equal to the critical speed v1, then the critical speed counter v1 is incremented by one. When the aforementioned energy-saving height counter h2 or default height counter h1 reaches the set value, the energy-saving height counter h2 or default height counter h1 is cleared to zero, and the visual inspection in step S3 is initiated. When the critical speed counter v1 reaches the set value, the energy-saving height counter h2, the default height counter h1 and the critical speed counter v1 in the default mode are all set to zero. (ii) In the passage mode, compare the vehicle speed with the critical speeds v3 and v4: If the vehicle speed is greater than the critical speed v4, the corresponding operation is to exit the pass mode and switch to the default mode, increment the exit counter v4, and reset the pass height counter h3 and the zero counter v34 to zero. If the vehicle speed is less than the critical speed v3, the corresponding operation is to adjust the vehicle height to the passing height, increment the passing height counter h3, and exit counter v4 and reset counter v34 to zero. If the vehicle speed is between the critical speeds v3 and v4, then the zero counter v34 is incremented by one; in: V3 is the dividing line between the vehicle's pass-through height and other heights; the vehicle is traveling at this speed indicating that it is attempting to overcome an obstacle. V4 is the threshold for entering and exiting the pass-through mode. Furthermore, v1 > v4 > v3. When the exit counter v4 or the pass height counter h3 reaches the set value, the counter is cleared to zero, and the visual inspection in step S3 is entered; when the zero counter v34 reaches the set value, the exit counter v4, the pass height counter h3, and the zero counter v34 in the pass mode are all cleared to zero. S3: Visual detection detects the environment in front of the vehicle; the data acquired by the camera is loaded into the trained YOLOv4-tiny traffic light recognition model for detection, and the height adjustment is determined based on the detection results. The following are two scenarios: 1) When a traffic light is detected ahead, maintain the altitude and do not perform any altitude switching, continue vehicle speed detection in this mode; however, an exit mechanism is established: in the default mode, if the vehicle speed is greater than the critical speed v1, return to the default mode to continue detection; in the pass mode, if the vehicle speed is greater than the critical speed v4, return to the default mode. 2) If no traffic light is detected, perform the corresponding height adjustment or mode switch based on the vehicle speed information; V4 is the threshold for entering and exiting pass mode; S4: If no traffic light is detected, adjust the height or switch modes; or if a traffic light is detected, keep the height unchanged and continue speed detection in this mode or switch to the default mode.

2. The computer vision-based vehicle height adjustment method according to claim 1, characterized in that: The YOLOv4-tiny traffic light recognition model was pre-trained twice using the large public dataset ImageNet and the public traffic light dataset WPI.

3. The computer vision-based vehicle height adjustment method according to claim 1, characterized in that: The data is self-collected, and the collected data includes images of traffic lights from multiple scenes, angles, and distances.

4. The computer vision-based vehicle height adjustment method according to claim 1, characterized in that: The self-collected dataset, after being labeled, is divided into a training set and a validation set in a 7:3 ratio.

Images