Target recognition-based car washing machine automatic control method and system

Abstract

The present application relates to the technical field of image data processing, in particular to a car washing machine automatic control method and system based on target recognition, comprising: acquiring continuous frame surface images of a vehicle to be washed at present and at multiple historical moments; calculating the stability degree, interference confidence and dynamic attention weight of a target pixel point, the target pixel point being any pixel point in the surface image corresponding to the vehicle to be washed at present; performing weighted processing on the feature map generated by the target recognition model in the feature extraction process based on the dynamic attention weight to obtain a weighted processed feature map, outputting the recognition result of stains of the vehicle to be washed based on the weighted processed feature map, and automatically washing the vehicle to be washed by using a water gun or a foam gun based on the recognition result. The present application solves the problem of low recognition accuracy of the prior art for washing control.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology. More specifically, this invention relates to an automatic control method and system for car wash machines based on target recognition. Background Technology

[0002] With the continuous development of the social economy and the rapid growth of private car ownership, automatic car wash equipment plays an increasingly important role in urban vehicle cleaning services. Traditional automatic car wash systems typically rely on preset programs and fixed spray paths, cleaning vehicle surfaces through uniform roller brushes, water guns, or foam sprays. They lack the ability to perceive and intelligently judge the distribution and type of dirt on the vehicle surface in real time. In practical applications, different vehicles have significantly different body structures, dirt types, and levels of contamination. For example, mud or dust may concentrate on the wheel arches, floorboards, or lower door edges; water stains and foam residue may appear on the roof or windshield edges; and stubborn stains may be randomly distributed. A uniform cleaning method under a fixed program cannot accurately handle these complex situations, often resulting in incomplete cleaning of some areas or requiring repeated cleaning to achieve the desired effect. This reduces car wash efficiency, increases water and detergent consumption, and affects the economic efficiency and service quality of automatic car wash equipment.

[0003] To improve the intelligence level of automatic car wash equipment, some existing technical solutions have introduced image recognition and target detection methods. These methods use industrial cameras to collect images of the vehicle surface and extract features from the images to identify areas of dirt.

[0004] However, most of these solutions rely on fixed feature maps in the feature extraction stage. That is, feature maps generated during model training or initialization are used for the analysis of all subsequent images. This method has significant technical limitations. Fixed feature maps cannot dynamically reflect subtle changes in the vehicle surface across consecutive frames and are easily affected by external factors such as lighting fluctuations, water spray disturbances, foam coverage, paint reflections, or slight vehicle vibrations. This leads to decreased target recognition accuracy and results in missed or false detections. Furthermore, fixed feature maps lack the ability to adaptively adjust to local pixel stability and interference levels, making it difficult to distinguish between real dirt and occasional noise. This results in insufficient stability of the recognition results in complex environments, leading to the problem of low accuracy in cleaning control recognition using existing technologies. Summary of the Invention

[0005] To address the problem of low accuracy in cleaning control and identification in the prior art mentioned in the background section, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides an automatic control method for a car wash machine based on target recognition, comprising: acquiring continuous frame surface images of the vehicle to be washed at the current time and multiple historical moments; calculating the stability of a target pixel, wherein the target pixel is any pixel in the surface image corresponding to the current moment of the vehicle to be washed; calculating the interference confidence of the target pixel, wherein the interference confidence is inversely correlated with the stability of the target pixel and the stability of each pixel within a set neighborhood centered on the target pixel, and positively correlated with the Euclidean distance between the target pixel and each pixel within the set neighborhood; calculating the dynamic attention weight of the target pixel, wherein the dynamic attention weight is inversely correlated with the interference confidence and positively correlated with the gradient magnitude of the target pixel; weighting the feature map generated by the target recognition model during feature extraction based on the dynamic attention weight to obtain a weighted feature map; outputting the recognition result of the stains on the vehicle to be washed based on the weighted feature map; and automatically washing the vehicle to be washed using a water gun or a foam gun based on the recognition result.

[0007] The above technical solution combines the pixel stability, neighborhood consistency, and structural saliency of continuous frame images, adaptively calculates the degree of pixel interference, and generates dynamic attention weights to weight the feature map of the target recognition model. This significantly enhances the ability to recognize real stains on the vehicle surface, while suppressing interference caused by changes in lighting, water flow disturbance, reflection, or foam coverage. It achieves accurate positioning of the stain area and directly guides the water gun or foam gun to perform intelligent, zoned, and differentiated cleaning. This not only improves the recognition accuracy and cleaning effect but also enhances resource utilization efficiency and the intelligence level of the automatic car wash system.

[0008] Furthermore, The corresponding pixel in the surface image at any time Stability for: , for The size of the time window for a given moment and its historical moments. for The corresponding pixel in the surface image at any time The gradient direction angle, for The corresponding pixel in the surface image at any time The gradient direction angle, For the natural constant An exponential function with base 0. for The corresponding pixel in the surface image at any time grayscale value, for Pixels in consecutive frames of surface images at multiple time points and their history The average gray level.

[0009] The above technical solution comprehensively evaluates pixel stability by combining the gradient direction consistency and grayscale change amplitude of pixels in consecutive multi-frame images. This enables it to highlight the continuous characteristics of the real surface structure while suppressing instantaneous noise and abnormal changes caused by light fluctuations, water flow disturbances, reflections, or foam obstruction. In terms of technical effect, it achieves adaptive screening of reliable pixels in consecutive frame images, providing more robust and accurate basic feature support for subsequent stain identification, anomaly detection, or cleaning control.

[0010] Furthermore, The corresponding pixel in the surface image at any time Interference confidence for: , for The corresponding pixel in the surface image at any time The degree of stability, For The time corresponds to the pixel in the surface image Pixels within the set neighborhood of the center The degree of stability, For the natural constant An exponential function with base 0. For pixels With pixels The Euclidean distance between them The total number of pixels within the defined neighborhood.

[0011] The above technical solution comprehensively evaluates the stability of a pixel itself and the overall consistency of its neighboring pixels, which can effectively quantify the possibility of a pixel being affected by external interference. This allows isolated unstable pixels or areas affected by factors such as water flow, reflection, and foam to be accurately identified. At the same time, it suppresses abnormal noise in stable local environments. In terms of technical effect, it achieves adaptive amplification of interference areas and retention of reliable pixels, providing a more robust foundation for subsequent dynamic weighted processing and accurate stain identification.

[0012] Furthermore, The corresponding pixel in the surface image at any time Dynamic attention weights for: , For the natural constant An exponential function with base 0. for The corresponding pixel in the surface image at any time The confidence level of the interference For normalization function, for The corresponding pixel in the surface image at any time The gradient magnitude.

[0013] The above technical solution combines the interference confidence of pixels with structural saliency and assigns dynamic attention weights to each pixel. This effectively suppresses the pixel weights in areas with large interference or instability, while pixels with clear edges and obvious structures can maintain high attention even with certain disturbances. Technically, this achieves adaptive enhancement of high-value, low-interference information in continuous frame surface images, improving the accuracy and robustness of subsequent feature extraction, stain recognition, and anomaly detection.

[0014] Furthermore, the weighted feature map is specifically generated as follows: in the feature extraction stage of the target recognition model, the original feature map generated by the target recognition model is obtained; for each pixel position in the original feature map, the dynamic attention weight of the corresponding pixel position is multiplied element-wise with the feature value of the corresponding pixel position to generate the weighted feature map.

[0015] The above technical solution achieves adaptive adjustment of the importance of feature information by multiplying the dynamic attention weights with the original feature map pixel by pixel. This strengthens stable and structurally significant regions in the feature map, while suppressing disturbed or unreliable regions. Technically, this improves the target recognition model's responsiveness to key targets and the effectiveness of feature representation, providing more accurate and robust basic features for subsequent stain detection or anomaly recognition.

[0016] Furthermore, the continuous frame surface images are acquired by an industrial camera mounted on the automated car wash equipment.

[0017] Furthermore, the target recognition model is the YOLO model.

[0018] Furthermore, the defined neighborhood range is 5. 5.

[0019] Furthermore, it also includes performing inter-frame alignment processing on the continuous frame surface images.

[0020] In a second aspect, the present invention provides an automatic control system for a car wash machine based on target recognition, including a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement the automatic control method for a car wash machine based on target recognition as described above.

[0021] The beneficial effects of this invention are as follows: This invention performs multi-dimensional analysis on continuous frame vehicle surface images, comprehensively considering pixel temporal stability, neighborhood consistency, and structural saliency. It adaptively calculates pixel interference levels and generates dynamic attention weights, which are then weighted during feature extraction in the target recognition model. This significantly enhances the feature representation of real stain areas while suppressing interference from light fluctuations, water flow disturbances, reflections, or foam coverage. This enables the system to accurately and robustly identify vehicle surface stains and intelligently control a water gun or foam gun for differentiated and targeted cleaning based on the recognition results. This not only improves the accuracy, efficiency, and reliability of stain detection and cleaning but also effectively saves water resources and cleaning agents, enhancing the intelligence level and application value of automatic car wash equipment. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an automatic control method for a car wash machine based on target recognition according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the comparison of the effects of the target recognition model before and after improvement in the automatic control method for car wash machines based on target recognition according to an embodiment of the present invention; Figure 3 This is a schematic block diagram illustrating the structure of an automatic control system for a car wash machine based on target recognition according to an embodiment of the present invention. Detailed Implementation

[0023] Example of an automatic control method for car wash machines based on target recognition.

[0024] like Figure 1 The flowchart shown below illustrates an embodiment of the automatic control method for a car wash machine based on target recognition, comprising the following steps: S1: Obtain continuous frame surface images of the vehicle to be cleaned at the current time and multiple historical moments.

[0025] In a preferred embodiment, the continuous frame surface images are acquired in real time by an industrial camera installed at a key station of the automatic car wash equipment. The industrial camera is preferably a high-resolution, high-frame-rate, and interference-resistant area-array or line-array industrial camera, which is rigidly mounted near the spray, brushing, or drying areas of the automatic car wash equipment to ensure stable and continuous acquisition of vehicle surface image data during vehicle operation. By appropriately setting the installation angle, focal length, and field of view of the industrial camera, the acquired images can completely cover the outer surface of the vehicle body, especially key areas prone to water stains, dirt, or residual foam, such as the doors, hood, roof, and rear, thus providing a sufficient and reliable visual information foundation for subsequent image analysis.

[0026] Furthermore, the industrial camera communicates with the control system or industrial IoT platform of the automatic car wash equipment to achieve synchronized triggering and acquisition with the car wash process. When a vehicle enters the predetermined detection area, the control system sends an acquisition command to the industrial camera. The industrial camera continuously acquires multiple frames of vehicle surface images according to a set sampling frequency, forming a temporally continuous surface image sequence. Further, the continuous frame surface images undergo inter-frame alignment processing. This continuous frame surface image not only reflects the changes in the cleanliness of the vehicle surface at different stages of the car wash but also effectively suppresses random noise introduced by light fluctuations, water flow disturbances, or slight vehicle vibrations in single-frame images, improving the stability and consistency of the overall image data.

[0027] S2: Calculate the stability, interference confidence, and dynamic attention weight of the target pixel, where the target pixel is any pixel in the surface image of the vehicle to be cleaned at the current moment.

[0028] In a preferred embodiment, The corresponding pixel in the surface image at any time Stability for: , for The size of the time window for a given moment and its historical moments. for The corresponding pixel in the surface image at any time The gradient direction angle, for The corresponding pixel in the surface image at any time The gradient direction angle, For the natural constant An exponential function with base 0. for The corresponding pixel in the surface image at any time grayscale value, for Pixels in consecutive frames of surface images at multiple time points and their history The average gray level.

[0029] The performance of the same pixel in multiple frames of surface images is comprehensively characterized from two complementary dimensions: temporal continuity and local consistency. On the one hand, by introducing a consistency measure of gradient direction change between consecutive frames, the structural stability of the pixel at different times is evaluated by using directional similarity. This enables a high level of stability when the real surface structure remains unchanged, while the stability is naturally weakened when the texture direction changes randomly due to water flow disturbance, foam occlusion, or noise interference. On the other hand, by statistically analyzing the grayscale fluctuation amplitude of the same pixel within a time window and suppressing it exponentially, the impact of unstructured brightness changes caused by illumination flicker, water droplet reflection, etc., on stability assessment can be effectively reduced, thus highlighting the persistent characteristics that reflect the true surface properties of the pixel. The two constraints are fused in a multiplicative manner and normalized in the time dimension, so that only pixels that simultaneously satisfy structural orientation stability and gradual grayscale changes can obtain a high degree of stability evaluation. This effectively suppresses the impact of transient noise, dynamic interference, and occasional anomalies on subsequent image analysis results, providing more robust and reliable basic feature support for surface state discrimination, cleanliness assessment, or anomaly detection.

[0030] The corresponding pixel in the surface image at any time Interference confidence for: , for The corresponding pixel in the surface image at any time The degree of stability, For The time corresponds to the pixel in the surface image Pixels within the set neighborhood of the center The degree of stability, For the natural constant An exponential function with base 0. For pixels With pixels The Euclidean distance between them. The defined neighborhood range is 5. 5. Of course, settings can also be made according to the actual situation. The total number of pixels within the defined neighborhood.

[0031] The temporal stability evaluation at the single pixel level is jointly modeled with the overall consistency of pixels in its spatial neighborhood to achieve a refined characterization of the possibility of pixels being disturbed by external factors: First, by inverting the stability of the pixel itself, the interference tendency of the pixel with more drastic fluctuations in the time dimension is actively amplified, thereby highlighting abnormal changes caused by factors such as water flow impact, foam blockage or instantaneous reflection. By further incorporating the stability of neighboring pixels and combining it with a weighting function that decays with spatial distance, the stable distribution of local regions surrounding a pixel is weighted and converged. This significantly suppresses interference confidence when the target pixel is in a generally stable local region, while simultaneously enhancing interference confidence when surrounding pixels exhibit unstable characteristics. By normalizing the neighborhood results and multiplicatively fusing them with the pixel's own instability, the phenomenon of local consistency disruption is effectively amplified. This allows for accurate differentiation between real surface structure changes and isolated noise points caused by random interference, thus enabling more reliable localization of disturbed regions in continuous frame surface image analysis. This provides a credible basis with both spatial and temporal constraints for subsequent anomaly identification, interference removal, or weight reallocation.

[0032] The corresponding pixel in the surface image at any time Dynamic attention weights for: , For the natural constant An exponential function with base 0. for The corresponding pixel in the surface image at any time The confidence level of the interference For normalization function, for The corresponding pixel in the surface image at any time The gradient magnitude.

[0033] By collaboratively modeling the credibility of pixel interference with the strength of its inherent structural information, dynamic adaptive adjustment of the attention level for different pixels is achieved. On the one hand, by using interference confidence as an exponential decay factor, the weight of pixels more likely to be affected by water flow disturbances, reflections, or noise in time and space is actively reduced, thus suppressing the interference of unreliable information on the overall analysis results. On the other hand, a normalized gradient magnitude is introduced as a representation of structural saliency, ensuring that areas with clear edges and distinct textures maintain high attention even under certain disturbances, while flat areas or areas with weak structural information are further weakened when interference is present. These two factors are embedded in an exponential function through a multiplicative relationship, resulting in a continuous and smooth nonlinear adjustment characteristic for weight changes. This achieves the prominent enhancement of high-value, low-interference pixels in continuous frame surface images and the adaptive suppression of low-value or high-interference pixels, effectively improving the sensitivity and overall robustness of subsequent feature extraction, state discrimination, or anomaly detection processes to key structural information.

[0034] S3: Based on the dynamic attention weights, the feature maps generated by the target recognition model during the feature extraction process are weighted to obtain the weighted feature maps.

[0035] In a preferred embodiment, the target recognition model is a YOLO model; the weighted feature map is specifically generated as follows: during the feature extraction stage of the target recognition model, the original feature map generated by the target recognition model is obtained; for each pixel position in the original feature map, the dynamic attention weight of the corresponding pixel position is multiplied element-wise with the feature value of the corresponding pixel position to generate the weighted feature map.

[0036] By introducing dynamic attention weights into the feature extraction process of the target recognition model, the importance of different pixels in the feature map is adaptively adjusted, significantly improving the model's responsiveness to key target regions and its ability to suppress interference regions. During the feature map generation stage, the original feature value of each pixel is weighted by its dynamic attention weight, preserving stable and structurally significant pixel information while reducing the influence of pixels affected by noise, reflection, or local anomalies. This pixel-by-pixel weighting process makes the weighted feature map more prominent in terms of target features and suppresses redundant or misleading information, thereby improving the accuracy, robustness, and ability to recognize small or low-contrast targets in complex environments. It also optimizes the model's stability and reliability in multi-frame continuous image processing, providing a solid data foundation for efficient and accurate target recognition.

[0037] S4: Based on the weighted feature map, output the identification result of the stains on the vehicle to be cleaned, and based on the identification result, use a water gun or foam gun to automatically clean the vehicle to be cleaned.

[0038] like Figure 2 The figure shown is a comparison of the effects of the target recognition model before and after improvement in the automatic control method for car wash machines based on target recognition according to an embodiment of the present invention.

[0039] In a preferred embodiment, the weighted feature map is used to accurately identify potential stain areas on the surface of the vehicle to be cleaned, and the corresponding stain distribution results are output. Specifically, after receiving the weighted feature map, the target recognition model can significantly enhance its focus on stable and structurally significant areas, while suppressing interference from noise, reflections, water droplets, or local abnormal pixels, thereby accurately locating stains on the vehicle surface, including various contaminants such as mud, dust, watermarks, or foam residue.

[0040] Based on the recognition results, the system can intelligently control a water gun or foam gun to spray different areas of dirt on the vehicle surface, prioritizing the cleaning of areas with high concentrations or high levels of contamination, thus achieving precise, zoned, and differentiated cleaning operations. By tightly coupling visual recognition with the automatic cleaning execution strategy, not only is the accuracy and real-time performance of dirt recognition improved, but also the waste of water and cleaning agents is avoided, increasing car wash efficiency and cleaning quality. Simultaneously, it enhances the adaptability of automatic car wash equipment to different types of contamination in complex environments, achieving an upgrade from passive cleaning to intelligent perception and proactive optimization cleaning, significantly enhancing the system's intelligence level and engineering application value.

[0041] The present invention combines pixel stability, neighborhood consistency, and structural saliency of continuous frame surface images to calculate the interference confidence of each pixel and generate dynamic attention weights. These weighted weights are then applied to the feature map of the target recognition model, thereby significantly enhancing the ability to identify real stains on vehicle surfaces. Simultaneously, it suppresses interference caused by changes in lighting, water flow disturbances, reflections, foam coverage, or local noise, achieving precise positioning of stain areas. Furthermore, it combines an automatic control device to intelligently spray water or foam guns, enabling differentiated and zoned cleaning. This not only improves the accuracy of stain detection and cleaning effect but also enhances resource utilization efficiency and the overall intelligence and robustness of the automatic car wash system.

[0042] Example of an automatic control system for car wash machines based on target recognition: like Figure 3 As shown in the figure, the structural block diagram of the automatic control system for a car wash machine based on target recognition according to an embodiment of the present invention includes a processor and a memory.

[0043] This invention also provides an automatic control system for car wash machines based on target recognition. For example... Figure 3 As shown, the system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the target recognition-based automatic control method for car wash machines according to the present invention.

[0044] The target recognition-based automatic control system for car wash machines also includes other components well-known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0045] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

[0046] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0047] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. An automatic control method for car wash machines based on target recognition, characterized in that, include: Acquire continuous frame surface images of the vehicle to be cleaned at the current time and multiple historical moments; The stability of the target pixel is calculated, where the target pixel is any pixel in the surface image of the vehicle to be cleaned at the current moment; the interference confidence of the target pixel is calculated, where the interference confidence is inversely correlated with the stability of the target pixel and the stability of each pixel in a set neighborhood centered on the target pixel, and positively correlated with the Euclidean distance between the target pixel and each pixel in the set neighborhood. Calculate the dynamic attention weight of the target pixel. The dynamic attention weight is inversely correlated with the interference confidence and positively correlated with the gradient magnitude of the target pixel. The feature map generated by the target recognition model during feature extraction is weighted based on the dynamic attention weight to obtain a weighted feature map. The recognition result of the stains on the vehicle to be cleaned is output based on the weighted feature map, and the vehicle to be cleaned is automatically cleaned using a water gun or foam gun based on the recognition result.

2. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, The corresponding pixel in the surface image at any time Stability for: , for The size of the time window for a given moment and its historical moments. for The corresponding pixel in the surface image at any time The gradient direction angle, for The corresponding pixel in the surface image at any time The gradient direction angle, For the natural constant An exponential function with base 0. for The corresponding pixel in the surface image at any time grayscale value, for Pixels in consecutive frames of surface images at multiple time points and their history The average gray level.

3. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, The corresponding pixel in the surface image at any time Interference confidence for: , for The corresponding pixel in the surface image at any time The degree of stability, For The time corresponds to the pixel in the surface image Pixels within the set neighborhood of the center The degree of stability, For the natural constant An exponential function with base 0. For pixels With pixels The Euclidean distance between them The total number of pixels within the defined neighborhood.

4. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, The corresponding pixel in the surface image at any time Dynamic attention weights for: , For the natural constant An exponential function with base 0. for The corresponding pixel in the surface image at any time The confidence level of the interference For normalization function, for The corresponding pixel in the surface image at any time The gradient magnitude.

5. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, The weighted feature map is specifically as follows: In the feature extraction stage of the target recognition model, the original feature map generated by the target recognition model is obtained; For each pixel position in the original feature map, the dynamic attention weight of the corresponding pixel position is multiplied element-wise with the feature value of the corresponding pixel position to generate a weighted feature map.

6. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, The continuous frame surface images are acquired by an industrial camera mounted on an automated car wash.

7. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, The target recognition model is the YOLO model.

8. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, The defined neighborhood range is 5.

5.

9. The automatic control method for a car wash machine based on target recognition according to claim 1, characterized in that, It also includes performing inter-frame alignment processing on the continuous frame surface images.

10. An automatic control system for a car wash machine based on target recognition, characterized in that, The system includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement the automatic control method for a car wash machine based on target recognition as described in any one of claims 1 to 9.

Images