A Method for Identifying the Working Area of ​​a Lawn-Mowing Robot Based on Multi-Feature Channel Quality Evaluation

Abstract

This invention discloses a method for identifying the working area of ​​a lawnmower robot based on multi-feature channel quality evaluation. Addressing the problem of unstable segmentation by a single fixed feature channel caused by lighting variations, noise interference, and motion blur in unstructured environments, this method constructs a subset of multi-feature channels consisting of the super-green index (ExG), HSV hue (H), and saturation (S). It calculates three-dimensional quantitative evaluation indicators for each feature channel, including inter-class separability, edge saliency, and environmental stability, and obtains a comprehensive quality score through weighted summation. The optimal feature channel is dynamically selected for Otsu threshold segmentation to obtain a preliminary binary image. Further image post-processing, including median filtering, morphological processing, contour filling, and connected component filtering, outputs a continuous and complete binary mask of the working area. This method offers high segmentation accuracy, strong anti-interference capability, requires no external facilities or labeled data, and has low computational complexity, making it suitable for real-time working area identification of lawnmower robots in complex outdoor environments.

Description

Technical Field

[0001] This invention belongs to the field of robot environmental perception and image processing technology, and relates to the visual recognition of the working area of ​​a lawnmower robot in an unstructured environment. Specifically, it is a method for identifying the working area of ​​a lawnmower robot based on multi-feature channel quality evaluation. Background Technology

[0002] During autonomous operation, lawnmower robots need to perceive their working environment and accurately distinguish between areas where lawns can be mowed and areas where they cannot be mowed, such as soil, rocks, debris, and roads. They must output complete and accurate boundary masks of the working areas in real time to plan their movement paths, avoid collision risks, ensure operational safety and mowing effectiveness, and achieve fully automated operation without human intervention. Currently, the mainstream solutions for obtaining lawn mowing area boundaries in the industry fall into two main categories: manually assisted deployment solutions and visual recognition solutions. However, both solutions have significant shortcomings and are difficult to adapt to complex and ever-changing unstructured outdoor environments.

[0003] Firstly, traditional physical / map boundary solutions, such as manual boundary setting, electronic fences, and pre-built maps, are costly to deploy and have poor environmental adaptability. Manual boundary setting requires operators to mark each courtyard on-site, which is extremely inefficient for large-area parks. Once the ground environment changes due to lawn growth, the addition of potted plants, leaf accumulation, or vegetation trimming, the original boundaries become invalid, requiring repeated on-site remarking, resulting in high labor costs in the long run. Electronic fences rely on pre-buried cables or signal transmitters, requiring a large amount of initial civil construction work. Pre-built map solutions rely on LiDAR and visual SLAM to pre-map the global environment, both of which have strict requirements for scene stability. If obstacles are added temporarily on-site, vegetation patterns change, or the area of ​​exposed soil changes, the stored map boundaries will have serious deviations, directly causing malfunctions such as missed or incorrect cutting of road obstacles.

[0004] Secondly, existing visual region segmentation algorithms for lawn mowing generally use a single fixed color feature channel to segment the lawn, which has inherent flaws in robustness. Current technologies often use the Super Green Index (ExG), HSV (Hyperhue or Saturation) single hue or saturation channel as the segmentation basis. Under ideal conditions of uniform lighting, no dust, and no motion blur, basic segmentation can be achieved. However, the actual outdoor working environment is subject to complex interference factors: backlighting in the early morning and late evening, tree shade, strong sunlight on cloudy days, dust noise, and motion blur caused by the robot's high-speed movement, all of which significantly weaken the distinguishability of lawn from the background using a single feature channel. When the lighting changes drastically, the grayscale distribution of the fixed feature channel will shift, and Otsu threshold segmentation is prone to large-area oversegmentation or undersegmentation, misclassifying bare soil as lawn or normal lawn as non-working areas. Fragmented segments, holes, and false targets are prevalent, making it impossible to output continuous and complete working area boundaries, severely restricting the autonomous operation capability of lawn mowing robots. Summary of the Invention

[0005] To address the technical problems existing in the background art, this invention proposes a method for identifying the working area of ​​a lawnmower robot based on multi-feature channel quality evaluation. This method can accurately identify the working area of ​​a lawnmower robot in unstructured environments, thereby improving the autonomous operation capability and adaptability of the lawnmower robot.

[0006] The objective of this invention can be achieved through the following technical solutions: A method for identifying the working area of ​​a lawnmower robot based on multi-feature channel quality evaluation includes the following steps: S1. Obtain the original color image of the working environment collected by the lawnmower robot, and extract feature channels of at least two different types of color components from the image to construct a multi-feature channel subset; S2 uses a preset three-dimensional quantitative evaluation index to calculate the comprehensive quality score of each feature channel in the multi-feature channel subset, and selects the feature channel with the largest comprehensive quality score as the optimal recognition channel to adapt to the current working environment. S3, perform binarization on the selected optimal recognition channel to obtain a preliminary binary segmentation image; S4 performs multi-level image post-processing operations on the initial binary segmentation image to eliminate image noise and false targets, and finally outputs a continuous and complete binary mask image of the working area.

[0007] Furthermore, the multi-feature channel subset in step S1 includes: the super green index ExG feature channel, the hue H component feature channel of the HSV color space, and the saturation S component feature channel of the HSV color space.

[0008] Furthermore, the three-dimensional quantization evaluation indicators preset in step S2 include: inter-class separability, edge saliency, and environmental stability. Inter-class separability is calculated based on the inter-class variance of the Otsu algorithm, edge saliency is calculated based on the gradient magnitude of the Sobel gradient operator and the proportion of strong pixels, and environmental stability is calculated based on the change rate of the Otsu threshold before and after adding Gaussian noise.

[0009] Further, in step S2, the three-dimensional quantitative evaluation index of each feature channel in the multi-feature channel subset is calculated, and the three-dimensional quantitative evaluation index is weighted and summed using preset weight coefficients to obtain the comprehensive quality score of multiple feature channels in sequence.

[0010] Furthermore, in step S3, the Otsu threshold segmentation algorithm is used to binarize the selected optimal recognition channel. In the resulting preliminary binary segmentation image, the foreground pixels correspond to the working area of ​​the lawnmower robot, and the background pixels correspond to the non-working area.

[0011] Furthermore, the multi-level image post-processing operations in step S4 include: median filtering, morphological processing, contour filling, and connected component filtering.

[0012] Furthermore, the morphological processing specifically includes: using elliptical structuring elements, first performing a closing operation on the filtered image to fill small holes, and then performing an opening operation to eliminate isolated small noise points.

[0013] Furthermore, the contour filling specifically includes: extracting the complete contour tree of the image using the RETR_CCOMP contour retrieval mode; filling all areas enclosed by the outer contours as foreground working areas; and for holes enclosed by the inner contours, performing the filling operation only when the hole area is less than a preset area threshold, while holes exceeding the threshold are retained as background non-working areas.

[0014] Furthermore, connected component filtering specifically includes: marking all independent connected components within the image, calculating the pixel area of ​​each connected component, retaining only connected components whose area accounts for a percentage greater than a preset threshold of the total image area, and removing small-area speckles and false targets.

[0015] The beneficial effects of this invention are: 1. This invention abandons the traditional approach of using a fixed single feature channel for recognition. Instead, it simultaneously constructs a feature subset consisting of three complementary feature channels: ExG super green index, HSV hue (H), and HSV saturation (S). This subset can compensate for the sensitivity to interference from lighting, soil, and shadows. Furthermore, it designs three-dimensional quantitative evaluation indicators for inter-class separability, edge saliency, and environmental stability. These indicators objectively quantify the image quality of each channel from three dimensions: ease of distinguishing foreground from background, clarity of lawn edges, and resistance to noise disturbance. A comprehensive score is then calculated using fixed weights to automatically select the optimal feature channel that best suits the lighting, motion blur, and dust noise conditions of the current frame. This effectively solves the problems of segmentation failure and foreground / background confusion in scenes with only ExG or HSV channels, such as backlighting, tree shading, and motion blur. Even in unstructured, variable outdoor environments, it can stably distinguish lawn work areas from unworkable areas such as soil, road surfaces, and debris. Its recognition adaptability and segmentation stability are significantly superior to traditional fixed feature algorithms.

[0016] 2. This invention employs multi-level image post-processing operations for binary segmented images. First, median filtering is used to smooth single-point noise. Then, elliptical structural elements are used to perform closing and opening operations sequentially to fill small holes and eliminate isolated micro-pollutants. Subsequently, layered filling based on the RETR_CCOMP contour tree is used to distinguish between large-area real holes and micro-holes, avoiding the mistaken filling of large non-working areas such as roads and flower beds as lawns. Finally, small-area false targets are eliminated by filtering based on the proportion of connected component areas. The entire process can simultaneously repair segmented and broken areas and eliminate false foregrounds formed by various interferences. The final output is a complete binary mask of the working area with continuous edges, no redundant holes, and no interference from speckles. It can directly provide reliable image basis for robot path planning and boundary determination, meeting the actual needs of outdoor lawn mowing projects.

[0017] 3. The overall algorithm of this invention is built entirely on classic image processing algorithms such as Otsu thresholding, Sobel gradient, morphological operations, and contour detection. It does not require pre-buried electronic fence cables, pre-built global maps with LiDAR, or semantic segmentation deep learning models. It does not require the collection and annotation of massive lawn image datasets for model training. The overall computational logic is simple, the computational load is low, and the hardware computing power requirements are relaxed. It can directly complete the real-time processing of single-frame images on the low-cost embedded main control chip built into the lawnmower robot without the need for high-performance computing units, effectively reducing the overall hardware cost of the robot. At the same time, it simplifies the on-site deployment process, eliminates the need for repeated manual boundary calibration, and greatly improves the efficiency of on-site equipment deployment. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0019] Figure 2 This is a schematic diagram of the three-dimensional quantitative evaluation and dynamic channel selection model of the present invention.

[0020] Figure 3 This is a schematic diagram of the multi-level image post-processing process of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figures 1-3 As shown, this invention provides a method for identifying the working area of ​​a lawnmower robot based on multi-feature channel quality evaluation, specifically including the following steps: S1. First, perform the environmental perception data acquisition step by continuously acquiring raw color images of the current working environment through the camera mounted on the lawnmower robot.

[0023] Then, based on the acquired original color image, the super green index ExG feature channel, the hue H component feature channel of the HSV color space, and the saturation S component feature channel of the HSV color space are extracted respectively to construct a multi-feature channel subset.

[0024] S2, for the current image, uses preset three-dimensional quantization evaluation metrics to calculate the comprehensive quality score of each feature channel in the multi-feature channel subset, and selects the feature channel with the largest comprehensive quality score as the optimal recognition channel for the current working environment. The preset three-dimensional quantization evaluation metrics include: inter-class separability, edge saliency, and environmental stability.

[0025] Specifically, the inter-class separability index is obtained by calculating the ratio of the inter-class variance of the target and background to the global variance using the Otsu algorithm. A higher ratio indicates that the target and background are more easily distinguished. The calculation formula is: ; in, and These represent the probabilities of the target and background pixels segmented by the Otsu algorithm, respectively. and This represents the average grayscale value for the corresponding category. The total variance of gray levels in the entire image. To prevent division by zero of constants.

[0026] Specifically, the edge saliency index is obtained by calculating the gradient magnitude using the Sobel operator and then statistically analyzing the proportion of strong pixels whose gradient magnitude is greater than the sum of the mean and standard deviation. The calculation formula is: ; ; ; in, and The horizontal and vertical gradients are calculated for the Sobel operator, respectively. and These are the mean and standard deviation of the gradient magnitude, respectively. The number of pixels with strong edges. This represents the total number of pixels in the image.

[0027] Specifically, the environmental stability index is obtained by adding Gaussian noise to the original channel and comparing the relative change in the Otsu threshold before and after the noise addition. The calculation formula is: ; ; in, The threshold is calculated using Otsu for the original channel. The threshold is calculated using Otsu after adding Gaussian noise to the original channel, with 10 in the denominator being an empirical constant.

[0028] The three-dimensional quantitative evaluation index of each feature channel in the multi-feature channel subset is calculated using the above formula. Then, the three-dimensional quantitative evaluation index is weighted and summed using preset weighting coefficients to obtain the comprehensive quality score for the three feature channels. The formula for calculating the comprehensive quality score is: ; Among them, the weighting coefficient satisfy In this embodiment, the weighting coefficient The values ​​are 0.5, 0.3, and 0.2.

[0029] S3. The selected optimal recognition channel is binarized using Otsu's threshold segmentation algorithm to obtain a preliminary binary segmentation image. In the preliminary binary segmentation image, the foreground pixels correspond to the working area of ​​the lawnmower robot, and the background pixels correspond to the non-working area.

[0030] S4 performs multi-level image post-processing operations on the initial binary segmentation image to eliminate image noise and false targets, ultimately outputting a continuous and complete binary mask image of the working region. These multi-level image post-processing operations include: median filtering, morphological processing, contour filling, and connected component filtering.

[0031] Morphological processing specifically includes: using elliptical structuring elements, first performing a closing operation on the filtered image to fill small holes, and then performing an opening operation to eliminate isolated small noise points. Contour filling specifically includes: extracting the complete contour tree of the image using the RETR_CCOMP contour retrieval mode; filling all areas enclosed by outer contours as foreground processing areas; for holes enclosed by inner contours, performing filling operations only when the hole area is less than a preset area threshold, and retaining holes exceeding the threshold as background non-processing areas. Connected component filtering specifically includes: marking all independent connected components in the image, calculating the pixel area of ​​each connected component, retaining only connected components whose area accounts for more than a preset percentage threshold of the total image area, and removing small-area noise and false targets.

[0032] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for identifying the working area of ​​a lawnmower robot based on multi-feature channel quality evaluation, characterized in that, Includes the following steps: S1. Obtain the original color image of the working environment collected by the lawnmower robot, and extract feature channels of at least two different types of color components from the image to construct a multi-feature channel subset; S2 uses a preset three-dimensional quantitative evaluation index to calculate the comprehensive quality score of each feature channel in the multi-feature channel subset, and selects the feature channel with the largest comprehensive quality score as the optimal recognition channel to adapt to the current working environment. S3, perform binarization on the selected optimal recognition channel to obtain a preliminary binary segmentation image; S4 performs multi-level image post-processing operations on the initial binary segmentation image to eliminate image noise and false targets, and finally outputs a continuous and complete binary mask image of the working area.

2. The identification method according to claim 1, characterized in that, The subset of multi-feature channels in step S1 includes: the super green index ExG feature channel, the hue H component feature channel of the HSV color space, and the saturation S component feature channel of the HSV color space.

3. The identification method according to claim 1, characterized in that, The preset three-dimensional quantization evaluation indicators in step S2 include: inter-class separability, edge saliency, and environmental stability. Inter-class separability is calculated based on the inter-class variance of the Otsu algorithm, edge saliency is calculated based on the gradient magnitude of the Sobel gradient operator and the proportion of strong pixels, and environmental stability is calculated based on the change rate of the Otsu threshold before and after adding Gaussian noise.

4. The identification method according to claim 3, characterized in that, In step S2, the three-dimensional quantitative evaluation index of each feature channel in the multi-feature channel subset is calculated, and the three-dimensional quantitative evaluation index is weighted and summed using preset weight coefficients to obtain the comprehensive quality score of multiple feature channels in turn.

5. The identification method according to claim 1, characterized in that, In step S3, the Otsu threshold segmentation algorithm is used to binarize the selected optimal recognition channel. In the resulting preliminary binary segmentation image, the foreground pixels correspond to the working area of ​​the lawnmower robot, and the background pixels correspond to the non-working area.

6. The identification method according to claim 1, characterized in that, The multi-level image post-processing operations in step S4 include: median filtering, morphological processing, contour filling, and connected component filtering.

7. The identification method according to claim 6, characterized in that, Morphological processing specifically includes: using elliptical structuring elements, first performing a closing operation on the filtered image to fill small holes, and then performing an opening operation to eliminate isolated small noise points.

8. The identification method according to claim 6, characterized in that, Contour filling specifically includes: extracting the complete contour tree of the image using the RETR_CCOMP contour retrieval mode; filling all areas enclosed by the outer contours as foreground working areas; and for holes enclosed by the inner contours, performing the filling operation only when the hole area is less than a preset area threshold, while holes exceeding the threshold are retained as background non-working areas.

9. The identification method according to claim 6, characterized in that, Connected component filtering specifically includes: marking all independent connected components within the image, calculating the pixel area of ​​each connected component, retaining only connected components whose area accounts for a percentage greater than a preset threshold of the total image area, and removing small-area speckles and false targets.

Images