Vehicle abnormal high temperature self-recognition and early warning method in free flow state

Abstract

The invention discloses a self-identification and early-warning method for abnormal high temperature of a vehicle in a free flow state, and the method achieves the precise identification of vehicle tire overheating and fire hazards through the fusion of thermal imaging analysis and a dynamic scene modeling technology. The method comprises the following steps: firstly, positioning a tire area in a thermal imaging image in real time by using a Yov11 model, segmenting an axle and a tread through a clustering algorithm in combination with temperature distribution and position characteristics, calculating a local average temperature and quantifying an abnormal index; meanwhile, environmental interference is eliminated by adopting a dynamic background modeling technology based on statistical analysis, efficient classification is performed on smoke areas in combination with a full convolutional neural network, and a smoke probability is output. Furthermore, the system dynamically evaluates the comprehensive risk level by fusing the temperature abnormal data and the smoke detection result, and links a grading early warning mechanism. According to the method, the detection precision and the response speed in a complex free flow scene are remarkably improved, the method can be widely applied to the fields of logistics transportation, dangerous goods vehicle monitoring and the like, and intelligent guarantee is provided for driving safety.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation monitoring technology, specifically a method for self-identification and early warning of abnormal high temperature in vehicles under free flow conditions. Background Technology

[0002] When vehicles are continuously moving in free-flow conditions, critical components such as tires and braking systems are prone to abnormally high temperatures due to factors such as overload, continuous operation, and high external temperatures, posing a significant safety risk. Current methods for identifying and warning of high temperatures in vehicles have obvious technical limitations.

[0003] First, existing methods mostly rely on fixed-location point-type infrared thermometers to measure moving vehicles, making it difficult to accurately capture the overall temperature distribution of key components, especially specific areas of the tires, when passing at high speeds. Measurement results are also susceptible to interference from angle, distance changes, and environmental heat sources, leading to insufficient representativeness of the temperature data. Second, most solutions are designed for static or low-speed scenarios, making them ill-suited for highly dynamic free-flowing environments such as highways. This significantly reduces the ability to stably locate fast-moving targets, accurately segment target components like tires, and extract temperatures, failing to guarantee reliable real-time and continuous monitoring. Furthermore, early warning mechanisms generally use a single temperature threshold, lacking dynamic consideration of vehicle operating conditions such as normal driving temperature rise and environmental factors such as road surface temperature. This makes it difficult to effectively distinguish between abnormally high temperatures and normal operating conditions, resulting in a persistently high false alarm rate. Summary of the Invention

[0004] To address the shortcomings in vehicle abnormal high temperature detection technology, this invention proposes a method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions.

[0005] The technical solution to achieve the purpose of this invention is: a method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions, comprising:

[0006] The thermal imaging camera collects real-time thermal imaging image data of the truck tires and the surrounding environment, and also acquires ambient temperature data.

[0007] The Yolov11 model is used to detect and locate tires in thermal imaging images in real time and crop out tire region sub-images.

[0008] Perform axle-tread region clustering segmentation on the cropped tire region sub-image;

[0009] Calculate the pixel-level average temperature of the axle and tire tread areas;

[0010] We use box plot statistical methods to remove moving interference targets from the background queue, construct a dynamic background model by the maximum and minimum values, and use the dynamic background model to achieve stable foreground detection in complex scenes.

[0011] A fully convolutional integral network is used to detect the probability of smoke presence in a foreground image.

[0012] The temperature anomaly is calculated by combining axle and tire tread temperatures, ambient temperature, and historical data.

[0013] Map the smoke detection probability to the smoke confidence level;

[0014] By integrating temperature anomaly and smoke confidence, a weighted comprehensive risk score is calculated, and the filtered comprehensive risk score is used to classify the early warning level.

[0015] Compared with the prior art, the significant advantages of this invention are:

[0016] (1) The dynamic risk assessment mechanism of multimodal data fusion breaks through the limitations of traditional single-indicator early warning. It integrates the tire temperature abnormality index and smoke confidence, and combines sliding window filtering to realize dynamic risk quantification, thus solving the problem of false alarms caused by environmental interference (such as clouds and fog being misjudged as smoke).

[0017] (2) A robust detection algorithm for complex scenes, based on box plot background modeling and a fully integral class network, can stably extract the foreground and identify slender smoke in dynamic interference (such as vehicle movement and leaf swaying), thus improving the speed and accuracy of smoke detection.

[0018] (3) Real-time graded response and closed-loop control strategy: low / medium / high level warnings are triggered by risk scoring, combined with continuous confirmation mechanism to suppress instantaneous false alarms, and linked vehicle braking and fire extinguishing systems, resulting in lower response delay and higher safety intervention efficiency.

[0019] Other features of the invention will be set forth in the following description, and will be apparent in part from the description. The objects of the invention can be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0020] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0021] Figure 1 This is a flowchart illustrating the technical process of the present invention.

[0022] Figure 2 This is a schematic diagram of the high-temperature anomaly detection facility of the present invention. Detailed Implementation

[0023] It is readily understood that, based on the technical solution of this invention, various embodiments of the invention can be conceived by those skilled in the art without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention. Rather, these embodiments are provided to enable those skilled in the art to gain a more thorough understanding of the invention. Preferred embodiments of the invention are described below in conjunction with the accompanying drawings, which form part of this application and, together with the embodiments of the invention, serve to illustrate the innovative concept of the invention.

[0024] A method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions, specifically including:

[0025] Step 1: Collect real-time thermal imaging image data of the truck tires and the surrounding environment using a thermal imaging camera, and simultaneously acquire ambient temperature data;

[0026] Step 2: Use the Yolov11 model to perform real-time detection and localization of tires in the thermal imaging image. The specific implementation process is as follows:

[0027] Input thermal imaging image (size H×W, temperature matrix T) raw (x,y)∈R H×W The image size was adjusted to Yolov11 standard input (640×640 pixels), while retaining the original temperature matrix T. raw ;

[0028] YOLOv11 outputs the tire bounding box coordinates (X... min ,X max ,Y min ,Y max );

[0029] Trim Tire Region Sub-image I tire Corresponding temperature matrix

[0030] Step 3: Perform axle-tread region clustering segmentation on the clipped tire region sub-image. The specific implementation process is as follows:

[0031] Build each pixel p i The three-dimensional feature vector [T] i ,x′ i ,y′ i ], where T i Let x′ be the temperature value at that point. i ,y′ i These are the normalized position coordinates of the point relative to the center of the tire region.

[0032] K-means clustering was used to divide all pixels into two categories: wheel axle and tire tread (k=1,2). Two pixels, μ1 and μ2, were randomly selected as the cluster centers of the two categories. The temperature features T1 and T2 of the two points were compared. The pixel with the higher temperature was selected as the cluster center of the wheel axle category (denoted as C1), and the pixel with the lower temperature was selected as the cluster center of the tire tread category (denoted as C2).

[0033] For each pixel p i Calculate the Euclidean distance between it and the current two cluster centers in the three-dimensional feature space:

[0034]

[0035] Point p i It will be assigned to the category C of the nearest cluster center. k For each category C k Calculate the average of the feature vectors of all pixels in this class, and use the two average points as the new cluster centers for the two classes. Repeatedly assign point p. i and update cluster center μ k The process continues until the cluster centers converge:

[0036] max k ||μ k (t+1) -μ k (t) ||<ε

[0037] Where t is the number of iterations, and ε is the convergence tolerance threshold (ε = 10). -3 ).

[0038] After clustering and classifying all pixels, morphological post-processing is performed to remove discrete noise points, retaining the largest connected region as the pivot R. hub With tread R tread .

[0039] Step 4: Calculate the pixel-level average temperature of the axle and tire tread areas:

[0040] Calculate the average temperature in the axle area. The specific implementation process is as follows:

[0041]

[0042] Calculate the average temperature of the tread area:

[0043]

[0044] Where, N hub and N tread This represents the number of pixels in the area corresponding to the axle and tire tread. and This represents the average temperature of the truck's wheel axle and tire tread.

[0045] Repeat the above steps for all g tires detected in the image, and output the value of each tire. and

[0046] Step 5: In the smoke detection stage, box plot statistical methods are used to remove moving interference targets from the background queue. A dynamic background model is constructed using maximum and minimum values ​​to achieve stable foreground detection in complex scenes, thereby detecting whether smoke is generated in the foreground. The specific implementation process is as follows:

[0047] To eliminate the interference of moving targets on foreground detection in free-flow scenes, a background queue B = {B1, B2, ..., B} is first established. m} to store historical context, where B i This represents the background image of the i-th frame, where m is the queue length;

[0048] To remove interfering pixels such as those with temporary obstructions, the historical value sequence is extracted for each pixel position (x, y) in the thermal imaging image.

[0049] V x,y ={B1(x,y),b2(x,y),…,B m (x,y)}

[0050] Calculate the upper and lower quartiles Q1(x,y) and Q3(x,y), and derive the interquartile range R. i0 (x,y) and outlier thresholds:

[0051] R i0 (x,y)=Q3(x,y)-Q1(x,y)

[0052] B L (x,y)=Q1(x,y)-1.5R i0 (x,y)

[0053] B U (x,y)=Q3(x,y)+1.5R i0 (x,y)

[0054] Remove V x,y China surpasses [B] L B U [Outliers], and use the median Fill in, generate a pure sequence

[0055] Further, pixel-level brightness extrema are extracted from the clean sequence to construct a dynamic background interval model, i.e., from Extract the maximum value V max(x,y) and the minimum value V min (x,y), construct a dynamic background model:

[0056] B model (x,y)=[V min (x,y),B max (x,y)]

[0057] The current frame thermal imaging image I t With B model (x,y) difference, extracting the foreground region F t :

[0058]

[0059] Finally, a frame-skipping interval n is set to update the background in real time. That is, every n frames, while keeping the background queue length m unchanged, the oldest frame (B1) in queue B is removed, and the original frames are moved forward in sequence. A new background frame is added to the end of the queue to ensure that the background update frequency covers the dwell time of the moving target and to avoid moving vehicles in free flow being mistakenly regarded as background.

[0060]

[0061] Where l represents the displacement (pixels) of the moving target in the image; v represents the moving speed (pixels / second); f represents the camera frame rate (frames / second); and m represents the background queue length (frames).

[0062] Step 6: To address the poor adaptability of traditional detection networks to elongated smoke, a fully convolutional network structure is designed, with the foreground F as the input. t The (x,y) image is processed through a convolutional layer Z. (k) =σ(W (k) *Z (k-1) +b (k) Multi-scale features are learned progressively. Among them, W... (k) and b (k) These represent the convolution kernel and bias of the k-th layer, respectively; σ represents the ReLU activation function; * represents the convolution operation, which progressively generates multi-scale feature maps Z. (k) The last convolutional layer outputs Z(L)(i,j);

[0063] The Z(L) is mapped to a spatial probability heatmap using the Sigmoid function:

[0064]

[0065] P smoke This represents the probability of smoke presence at each position (i,j), enabling end-to-end smoke recognition.

[0066] Step 7: Combining axle and tire tread temperatures, ambient temperature, and historical data, define a temperature anomaly index (α) to determine temperature anomalies:

[0067] The temperature of each tire on the truck is assessed one by one, and the temperature anomaly α of the g-th tire is defined. g :

[0068]

[0069] Where T thr T is the empirical threshold temperature. env The ambient temperature can be directly measured by an infrared thermal imager. Let g be the average temperature of the g-th tire axle region. Let be the average temperature of the g-th tire tread region.

[0070] Step 8: Map the smoke detection probability to a confidence index (β), and calibrate the probability interval using a piecewise function to avoid interference from low confidence levels;

[0071] Define the smoke confidence level β:

[0072]

[0073] Step 9: Combine temperature anomaly (α) and smoke confidence (β) to calculate a weighted comprehensive risk score (S):

[0074] Define the overall risk score as S:

[0075] S=ω1·α g +ω2·β

[0076] Where ω1+ω2=1;

[0077] To suppress false triggering caused by instantaneous fluctuations, a time series analysis step is introduced, and a moving average filter is applied to S. At time t, the filtered risk score is:

[0078]

[0079] Where N is the window length, in frames;

[0080] According to S filtered Classification of warning levels:

[0081]

[0082] M is set according to the actual application scenario. In addition, if β = 1 or α g ≥1.5, high-risk warnings are triggered independently in emergency situations.

[0083] The system displays real-time warning levels, abnormal tire locations, and temperature distribution cloud maps, and uploads the warning information to the monitoring center. In low-risk situations, roadside LED warnings are triggered to stop the vehicle; in medium-risk situations, audible and visual alarms are activated; and in high-risk situations, the vehicle is blocked off-road with water to cool it down.

[0084] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

[0085] It should be understood that, in order to simplify the present invention and help those skilled in the art understand its various aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes described in a single embodiment or with reference to a single figure. However, the present invention should not be construed as including all features in the exemplary embodiments as essential technical features of the claims of this patent.

[0086] It should be understood that the modules, units, components, etc., included in the device of one embodiment of the present invention can be adaptively changed to be placed in a device different from that embodiment. Different modules, units, or components included in the device of the embodiment can be combined into a single module, unit, or component, or they can be divided into multiple sub-modules, sub-units, or sub-components.

Claims

1. A method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions, characterized in that, include: The thermal imaging camera collects real-time thermal imaging image data of the truck tires and the surrounding environment, and also acquires ambient temperature data. The Yolov11 model is used to detect and locate tires in thermal imaging images in real time and crop out tire region sub-images. Perform axle-tread region clustering segmentation on the cropped tire region sub-image; Calculate the pixel-level average temperature of the axle and tire tread areas; We use box plot statistical methods to remove moving interference targets from the background queue, construct a dynamic background model by the maximum and minimum values, and use the dynamic background model to achieve stable foreground detection in complex scenes. A fully convolutional integral network is used to detect the probability of smoke presence in a foreground image. The temperature anomaly is calculated by combining axle and tire tread temperatures, ambient temperature, and historical data. Map the smoke detection probability to the smoke confidence level; By integrating temperature anomaly and smoke confidence, a weighted comprehensive risk score is calculated, and the filtered comprehensive risk score is used to classify the early warning level.

2. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 1, characterized in that, The specific method for performing axle-tread region clustering segmentation on the clipped tire region sub-image is as follows: Build each pixel p i The three-dimensional feature vector [T] i ,x′ i ,y′ i ], where T i Let x′ be the temperature value at that point. i ,y′ i The normalized position coordinates of this point relative to the center of the tire region. All pixels are divided into two categories: wheel axle and tire tread. Two pixels, μ1 and μ2, are randomly selected as cluster centers for the two categories. The temperature features T1 and T2 of the two pixels are compared. The pixel with the higher temperature is selected as the cluster center C1 for the wheel axle category, and the pixel with the lower temperature is selected as the cluster center C2 for the tire tread category. Calculate each pixel p i Euclidean distance between the current two cluster centers and the two cluster centers in the three-dimensional feature space: Point p i It was assigned to the category C of the nearest cluster center. k Calculate C for each category k The average value of the feature vectors of all pixels is used to calculate the two average values, which are then used as the new cluster centers for the two categories. Repeated assignment point p i and update cluster center μ k The process continues until the cluster centers converge; After clustering and classifying all pixels, morphological post-processing is performed to remove discrete noise points, retaining the largest connected region as the pivot R. hub With tread R tread .

3. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 1, characterized in that, The specific formulas for calculating the pixel-level average temperature of the axle and tire tread areas are as follows: Calculate the average temperature in the wheel axle area: Calculate the average temperature of the tread area: Where, N hub and N tread This represents the number of pixels in the area corresponding to the axle and tire tread. and R represents the average temperature of the truck's axle and tire tread. hub For the wheel and axle region, R tread This refers to the tire tread area.

4. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 1, characterized in that, The specific method for using box plot statistical methods to remove moving interference targets from the background queue and constructing a dynamic background model based on the maximum and minimum values ​​is as follows: Establish a background queue B = {B1, B2, ..., B} m } to store historical context, where B i This represents the background image of the i-th frame, where m is the queue length; Extract the historical value sequence of each pixel position (x, y) in the thermal imaging image: V x,y ={B1(x,y),B2(x,y),…,B m (x,y)} Calculate the upper and lower quartiles Q1(x,y) and Q3(x,y), and derive the interquartile range R. i0 (x,y) and outlier thresholds: R i0 (x,y)=Q3(x,y)-Q1(x,y) B L (x,y)=Q1(x,y)-1.5R i0 (x,y) B U (x,y)=Q3(x,y)+1.5R i0 (x,y) Remove V x,y China surpasses [B] L B U [Outliers], and use the median Fill in, generate a pure sequence A dynamic background interval model is constructed by extracting pixel-level brightness extrema from a clean sequence, i.e., from... Extract the maximum value V max (x,y) and the minimum value V min (x,y), construct a dynamic background model: B model (x,y)=[V min (x,y),V max (x,y)] The current frame thermal imaging image I t With B model (x,y) difference, extracting the foreground region F t : Set a frame interval n to update the background in real time. That is, every n frames, while keeping the length m of the background queue unchanged, remove the oldest frame in queue B, move the original frames forward in turn, and add a new background frame to the end of the queue. This ensures that the background update frequency covers the dwell time of the moving target and avoids moving vehicles in free flow state being mistakenly regarded as background.

5. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 4, characterized in that, The specific frame extraction interval n is: In the formula, l represents the displacement of the moving target in the image; v represents the moving speed; f represents the camera frame rate; and m represents the background queue length.

6. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 1, characterized in that, The specific method for detecting the probability of smoke presence in a foreground image using a fully convolutional integral network is as follows: Input foreground F t The (x,y) image is processed through a convolutional layer Z. (k) =σ(W (k) *Z (k-1) +b (k) (Learning multi-scale features stepwise, where W) (k) and b (k) These represent the convolution kernel and bias of the k-th layer, respectively; σ represents the ReLU activation function; * represents the convolution operation, which progressively generates multi-scale feature maps Z. (k) The last convolutional layer outputs Z(L)(i,j); The Z(L) is mapped to a spatial probability heatmap using the Sigmoid function: P smoke This represents the probability of smoke being present at each position (i,j).

7. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 1, characterized in that, The formula for calculating the degree of temperature anomaly is: Where T thr T is the empirical threshold temperature. env The ambient temperature can be directly measured by an infrared thermal imager. Let g be the average temperature of the g-th tire axle region. Let be the average temperature of the g-th tire tread region.

8. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 1, characterized in that smoke... The formula for calculating confidence level is: In the formula, P smoke The probability of smoke being present at each position (i,j).

9. The method for self-identification and early warning of abnormal high temperature in vehicles under free-flow conditions according to claim 1, characterized in that, The specific method for classifying early warning levels by integrating temperature anomaly and smoke confidence, weighting the comprehensive risk score, and then using the filtered comprehensive risk score is as follows: The overall risk score is calculated as S: S=ω1·α g +ω2·β Where ω1+ω2=1; Applying a moving average filter to the comprehensive risk score, the filtered risk score at time t is: Where N is the window length; According to S filtered Classification of warning levels: If β = 1 or α g ≥1.5, high-risk warning is triggered independently in emergency situations, where M is the set value.

Images