CN-121982655-A - Method and device for monitoring boiling state of liquid in cooker

CN121982655ACN 121982655 ACN121982655 ACN 121982655ACN-121982655-A

Abstract

The invention provides a method and a device for monitoring a liquid boiling state in a pot, which comprise the steps of collecting an empty pot image, sequentially carrying out pretreatment, pot shape identification, ROI area determination, ROI verification and pot edge height detection based on the empty pot image, carrying out background modeling and foreground extraction based on the empty pot image, determining a real boiling area of the empty pot image based on various characteristics, sequentially carrying out mass center height calculation, high point height calculation, height data smoothing and boiling energy index calculation based on the real boiling area of the empty pot image, determining a monitoring state based on the boiling energy index and the height data, and determining a monitoring strategy based on the monitoring state, wherein the monitoring state comprises a silence period, an active period and a critical period. The non-contact, high-precision, dynamic and self-adaptive monitoring and early warning of the boiling state of the liquid of the pot can be realized, the core problems of manual dependence, static processing defects and early warning hysteresis in the prior art are solved, and the intelligent liquid boiling state monitoring device can be seamlessly integrated in kitchen equipment such as intelligent stoves.

Inventors

WANG LIANGBAO
Jiang Lianzeng
CHENG XIAOLEI
REN FUJIA

Assignees

杭州老板电器股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. A method for monitoring the boiling state of a liquid in a pot, the method comprising: Acquiring an empty pot image, and sequentially carrying out pretreatment, pot shape recognition, ROI region determination, ROI verification and pot edge height detection based on the empty pot image; Performing background modeling and foreground extraction based on the empty pan image, and determining a real boiling region of the empty pan image based on various features; sequentially performing centroid height calculation, high point height calculation, height data smoothing and boiling energy index calculation based on the real boiling region of the empty pot image; a monitoring state is determined based on the boiling energy index and the height data, and a monitoring strategy is determined based on the monitoring state, wherein the monitoring state comprises a silence period, an active period and a critical period.
2. The method of claim 1, wherein the steps of preprocessing, pan shape recognition, ROI area determination, ROI verification, and rim height detection are sequentially performed based on the empty pan image, comprising: sequentially carrying out graying, gaussian blur processing and Canny edge detection pretreatment on the empty pot image; identifying the shape of the pot by means of Hough transformation on the empty pot image; Determining a detection area covering liquid in the pot as an ROI area based on the shape of the pot; Calculating gray variance of the ROI area; if the gray variance is not in the threshold range, determining that the ROI verification is not passed; if the ROI verification is passed, detecting the pan edge height of the pair of empty pan images by means of vertical gradient analysis in combination with curve fitting.
3. The method of claim 1, wherein the steps of background modeling and foreground extraction based on the empty pot image comprise: performing background modeling on the empty pot image through a Gaussian mixture model; designing a foreground counter for each pixel of the empty pot image; Calculating a dynamic learning rate based on the foreground counter; Updating the background model of the non-foreground region; shadow suppression is performed based on the color luminance difference and the texture feature.
4. The method of claim 1, wherein determining a true boiling region of the empty pan image based on a plurality of features comprises: Determining a foreground pixel block of the empty pan image; verifying the foreground pixel blocks based on geometric features, texture features and motion consistency features respectively; If the verification based on the geometric feature, the texture feature and the motion consistency feature is all passed, determining the foreground pixel block as a real boiling region of the empty pan image; if at least one of the verification based on the geometric feature, the texture feature, and the motion consistency feature fails, determining that the foreground pixel block is not a true boiling region of the empty pot image.
5. The method of claim 1, wherein the steps of centroid height calculation, high point height calculation, height data smoothing and boiling energy index calculation are sequentially performed based on the real boiling region of the empty pan image, comprising: calculating the vertical height of the mass center of the real boiling region of the empty pan image in a pixel coordinate weighted average mode; detecting the highest pixel point of a real boiling region in the empty pot image in a single frame as a high point height; Carrying out the height data smoothing treatment in a multi-frame moving average mode; and calculating the boiling energy index based on the boiling region area ratio and the bubble motion speed of the empty pot image.
6. The method of claim 1, wherein the step of determining a monitoring policy based on the monitoring status comprises: If the monitoring state is a silent period, reducing the image acquisition frame rate, closing part of the feature analysis module, and reducing the CPU calculation load; if the monitoring state is an active period, adopting a standard image acquisition frame rate, and starting all feature analysis modules; And if the monitoring state is a critical period, increasing the image acquisition frame rate, starting predictive overflow early warning, and increasing the height data sampling frequency.
7. The method of claim 6, wherein initiating a predictive overflow alert comprises: acquiring historical height data and a historical boiling energy index, and preprocessing the historical height data and the historical boiling energy index; Performing linear extrapolation prediction on the historical height data and the historical boiling energy index to determine future height data and future boiling energy; Determining a risk level based on the future altitude data and the future boiling energy; And carrying out predictive overflow early warning based on the risk level.
8. The method according to any one of claims 1-7, further comprising: Starting an automatic calibration flow, prompting a user to add liquid with normal cooking quantity, heating the liquid to boil according to a normal mode after the liquid is added, and recording the height data and the boiling energy index; adjusting a threshold based on the recorded height data and the boiling energy index; And detecting the overflow of the adjusted threshold, and determining whether the adjusted threshold is accurate or not based on the false alarm rate and the missing alarm rate of the overflow detection.
9. The method according to any one of claims 1-7, further comprising: Recording key data of each cooking, and establishing a cooking habit database based on the key data, wherein the key data comprises a cooker type, a liquid type, heating power, early warning times and user intervention behaviors; and/or determining early warning accuracy of each cooking scene based on the cooking history data, and optimizing a threshold value of the corresponding cooking scene based on the early warning accuracy.
10. A device for monitoring the boiling state of a liquid in a pan, the device comprising: The initialization stage module is used for acquiring an empty pot image, and sequentially carrying out pretreatment, pot shape recognition, ROI region determination, ROI verification and pot edge height detection based on the empty pot image; The real-time monitoring stage module is used for carrying out background modeling and foreground extraction based on the empty pot image and determining a real boiling region of the empty pot image based on various characteristics; The quantitative analysis stage module is used for sequentially carrying out centroid height calculation, high point height calculation, height data smoothing and boiling energy index calculation based on the real boiling region of the empty pot image; And the dynamic adjustment stage module is used for determining a monitoring state based on the boiling energy index and the height data and determining a monitoring strategy based on the monitoring state, wherein the monitoring state comprises a silence period, an active period and a critical period.

Description

Method and device for monitoring boiling state of liquid in cooker Technical Field The invention relates to the technical field of intelligent home furnishing, in particular to a method and a device for monitoring the boiling state of liquid in a pot. Background In a kitchen cooking scene, monitoring of the boiling state of liquid in a pot is a core link for guaranteeing the cooking effect and the use safety. The prior art scheme has the following key limitations: 1. The traditional scheme requires continuous observation by users, can not liberate hands, is easy to overflow liquid, dry burn and even fire caused by negligence, and has poor suitability especially for the old people; 2. The robustness of the traditional sensor is insufficient, the temperature sensor is influenced by the difference of boiling points of liquid (such as 108 ℃ of boiling point of brine and 99 ℃ of pure water) and the thermal conductivity of the material of a cooker (stainless steel vs ceramic), the misjudgment rate is more than 15%, the sound sensor is easily interfered by environmental noise such as a range hood (60-70 dB), vegetable cutting sound (55 dB) and the like, and the effective recognition distance is less than 1 meter; 3. The static defect of the computer vision scheme is that the existing vision monitoring mostly adopts a fixed threshold value and single foreground extraction (such as a simple frame difference method), cannot adapt to the dynamic characteristics of the boiling process (such as gradual change from micro boiling to severe boiling), has a false judgment rate of 8% on steam reflection (kitchen humidity >60% time-frequency) and instantaneous splashing (such as porridge boiling bubbling), and lacks the self-adaptation capability of long-term use scenes (such as manual calibration after a user changes cookware). Disclosure of Invention In view of the above, the present invention aims to provide a method and a device for monitoring the boiling state of a liquid in a pot, so as to realize a closed loop of "dynamic sensing-intelligent decision-prediction early warning" through the depth optimization of a traditional algorithm (independent neural network), keep high-efficiency operation on embedded equipment (single frame processing <5 ms), and achieve the sensing precision close to an intelligent model. According to the method, pretreatment, cooker shape identification, ROI area determination, ROI verification and pot edge height detection are sequentially conducted on the basis of empty cooker images, background modeling and foreground extraction are conducted on the basis of the empty cooker images, real boiling areas of the empty cooker images are determined on the basis of various features, centroid height calculation, high point height calculation, height data smoothing and boiling energy index calculation are sequentially conducted on the basis of the real boiling areas of the empty cooker images, a monitoring state is determined on the basis of the boiling energy index and the height data, and a monitoring strategy is determined on the basis of the monitoring state, wherein the monitoring state comprises a silence period, an active period and a critical period. In an alternative embodiment of the application, the steps of preprocessing, pan shape recognition, ROI area determination, ROI verification and pot edge height detection based on the empty pot image sequentially comprise the steps of performing graying, gaussian blur processing and Canny edge detection preprocessing on the empty pot image sequentially, recognizing the pan shape of the empty pot image in a Hough transformation mode, determining a detection area covering liquid in the pot as an ROI area based on the pan shape, calculating gray variance of the ROI area, determining that the ROI verification is passed if the gray variance is within a preset threshold range, determining that the ROI verification is not passed if the gray variance is not within the threshold range, and detecting the pot edge height of the empty pot image in a mode of combining vertical gradient analysis with curve fitting if the ROI verification is passed. In an alternative embodiment of the application, the steps of background modeling and foreground extraction based on the empty pan image comprise the steps of carrying out background modeling on the empty pan image through a Gaussian mixture model, designing a foreground counter for each pixel of the empty pan image, calculating a dynamic learning rate based on the foreground counter, carrying out background model updating on a non-foreground area, and carrying out shadow suppression based on color brightness difference and texture features. In an alternative embodiment of the present application, the step of determining the real boiling area of the empty pan image based on the multiple features includes determining a foreground pixel block of the empty pan image, verifying the foreground pixel block based on the geom