46results about How to "Low image quality requirements" patented technology

The invention discloses a motion unit layering-based facial expression recognition method and system. The facial expression recognition method comprises steps of classifying three layers, and specifically comprises the following steps: first extracting an area adjacent to the upper part of the nose as a first-layer classification area, and roughly classifying an expression by taking whether an AU9 motion unit is detected or not as a judgment standard of a first-layer classifier; then extracting a lip area as a second-layer classification area, and performing fine adjustment on the basis of a first-layer classification result by taking whether AU25 and AU12 motion units are detected or not as a judgment standard of a second-layer classifier; finally extracting an upper half face area and a lower half face area as third-layer classification areas respectively, and performing precision classification on the basis of a second-layer classification result. The invention further provides the system for implementing the method. According to the method and the system, characteristics of representative areas of the expression are extracted on the basis of an AU layered structure, and layer-by-layer random forest classification is combined, so that expression recognition accuracy is effectively improved, expression recognition speed is increased, and the method and the system are particularly applied to a low-resolution image.

The invention discloses a robot vision locating method based on round road sign imaging analysis, belonging to the technical field of robot vision location. The robot vision locating method is characterized by comprising the following steps of: designing and placing a circular road sign; identifying the circular road sign, wherein the step of identifying the circular road sign comprises the three procedures of fitting edges with specific colors, fitting RANSAC (Random Sampling Consensus) ellipse and checking the sign; and finally, locating a mobile robot. The robot vision locating method has the advantages of location completion by using only one road sign, simple road sign mode, short detection process, capability of being used in outdoor light conditions and simpleness and rapidness in location.

The invention relates to a pointer instrument detection and reading identification method based on a mobile robot. The method comprises the following steps: obtaining a deep neural network detection model M for a pointer instrument; the mobile robot moves to a designated place to obtain an original environment image containing the instrument equipment at present in a mode that the mobile robot carries a camera; S serves as system input and is transmitted to a deep neural network model M, whether an instrument exists in S or not is detected, the position of the instrument is framed out, an image in a frame is intercepted and subjected to height setting processing, the length-width ratio is not changed, and the processing result is represented by J; contrast enhancement is performed on the image, and a processing result is represented by E; local adaptive threshold segmentation is performed on the E to obtain a reverse binary image B; pointer extraction processing is performed based on aprobability circle; a central straight line L of the pointer part is extracted as pointing information of the instrument pointer; and a coordinate system is established based on a probability circlecenter projection algorithm and reading by an angle method.

Disclosed are a method, apparatus and electronic device for detecting dirty areas in an image. The method includes: performing a first scale Gaussian filter on an image to be detected to obtain a background image corresponding to the image to be detected; Making a difference between the image to be detected and the background image to obtain a difference image; Performing a second scale Gaussian filter on the difference image to obtain a resultant image, the scale of the second scale Gaussian filter being smaller than the scale of the first scale Gaussian filter; And determining a dirty area in the image based on the resultant image. In this way, the sensitivity, ease of use, processing speed and environmental adaptability of dirt detection can be improved.

The invention provides a method for detecting a bin level in an image processing mode based on a digital camera and auxiliary lighting equipment. The method provided by the invention comprises the following steps of: grading the range of a bin by summarizing the characteristics of an image environment of the bin and analyzing the distribution law of the image entropy of defocused images within the full range of the bin in the environment of the bin; controlling the lens focus of the digital camera by a program so that the camera captures images based on grades; preprocessing the images firstly, and extracting feature textures of materials; and calculating the image entropy of the processed images, and finally obtaining a bin level value through analysis and calculation.

Embodiments of the present invention provide a license plate recognition method, device and system, and a storage medium. The method comprises the steps: obtaining an image to be subjected to licenseplate recognition; performing character recognition on the image to be subjected to license plate recognition, so as to obtain a preliminary recognition result of the character; and determining a final recognition result based on license plate prior information and the preliminary recognition result. The license plate prior information is utilized for determining the final recognition result, andthe serious dependence of the license plate recognition technology on the image quality can be reduced according to the technical scheme. Even if the quality of the image to be subjected to license plate recognition is poor, an ideal recognition result can still be obtained based on the license plate prior information, thereby the license plate identification precision is remarkably improved.

The invention discloses a protection pressing plate state identification method based on image processing shadow removal optimization, and the method comprises the steps: carrying out the graying of aprotection pressing plate color image, converting the protection pressing plate color image into a gray-scale image, carrying out the contrast enhancement and binarization of the gray-scale image, and eliminating a shadow region; obtaining a convex hull of each protection pressing plate switch through a Graham algorithm principle, then connecting the convex hulls into a rectangle through a minimum enclosing rectangle principle, and obtaining the rectangular area. setting a threshold value for the rectangular area; if the rectangular area is greater than the threshold value, determining that throwing-out is performed, otherwise, determining that throwing-in is performed. According to the method, the influence of shadow interference can be effectively reduced, the requirement on the qualityof the acquired image is not high, and the robustness is high; the influence of shadow interference can be effectively reduced, and the operation state of the pressing plate in the image can be accurately identified.

The embodiment of the invention discloses a meter reading method based on image recognition. The method comprises the following steps: acquiring a meter image of an electric energy meter; carrying out homogeneity conversion on the meter image to obtain a homogeneity image which is a single-channel image with the same characteristics of white reading numbers and dark color backgrounds; determining a reading area image from the homogeneity image; obtaining a character area image according to the reading area image; and identifying each character in the character area image to obtain reading information of the electric energy meter.

The invention provides a data encoding method and device, and relates to the technical field of electronic information, capable of solving the problem of resource waste during encoding processing of image frames. The specific technical scheme includes the steps: generating a target image frame sequence after a display image of terminal equipment is acquired; calculating the change frequency corresponding to the target image frame sequence according to a preset algorithm; and according to the change frequency, when the change frequency is greater than a preset value, performing encoding processing on the target image frame sequence according to a first encoding rule, and when the change frequency is less than or equal to the preset value, performing encoding processing on the target image frame sequence according to a second encoding rule. The present disclosure is used for image encoding.

The invention discloses a squid freshness identification method based on color space transformation and pixel clustering, and the method comprises the steps: unfreezing squids, cleaning the unfrozen squids, and preparing squid samples; unfolding the squid sample on a workbench, placing the squid sample in an auxiliary light source irradiation area, and performing image acquisition on the squid sample at different angles by using shooting equipment to obtain an original squid image; carrying out image preprocessing to obtain a test image; carrying out color space transformation and pixel clustering on the test image, extracting a red decay area of the test image, carrying out ratio calculation on the red decay area and the total surface area of the squid, dynamically analyzing the meat quality change condition of the squid, and monitoring the decay rate of the squid. According to the method, the image processing technology is utilized, color space transformation and pixel clustering analysis are carried out on the shot squid images to obtain the metamorphic region area of the squid images, ratio calculation is carried out on the metamorphic region area and the total surface area, and therefore non-contact and lossless squid freshness dynamic monitoring and identification under different storage durations and temperatures are achieved.

The invention belongs to the technical field of disease and insect pest identification, and discloses a crop disease and insect pest intelligent identification method. The method comprises the steps:S1, setting a region range, and obtaining basic information of crops in multiple time periods in the region range; S2, acquiring image data and meteorological data in a regional range; S3, acquiring first feature information reflecting the physiological features of the crops from the image data; S4, judging whether the physiological characteristics of the crops in the area range are normal or notin the current time period, and if yes, ending the process; if not, entering the step S5; S5, identifying a suspected area with abnormal physiological features, and generating a crop resource image; S6, acquiring second feature information reflecting the physiological features of the crops; S7, comparing and judging whether the physiological characteristics of the crops in the suspected area are normal or not, and if yes, ending the process; if not, entering the step S8; and S8, making the final judgment manually. The method is low in cost and accurate in judgment.

The invention discloses an intelligent collection method of machine room equipment display panel data, which comprises the following steps: S1, a robot acquires images of a plurality of display panelsas a training data set; s2, inputting the training data set into an improved master-rcnn algorithm, and training to obtain a text detection model; s3, the robot collects an image of the display panelin real time and inputs the image into the text detection model obtained through training in the step S2, all texts are automatically marked, so that detection boxes are obtained, and position coordinates and size information of all the detection boxes in the image in the display panel are output; s4, extracting a Roi image in the detection frame, performing image preprocessing on the Roi image,and reserving the extracted digital skeleton image as a training sample set; s5, training an svm classifier according to the training sample set, and performing classification recognition on a singlenumber through the svm classifier; and S6, splicing the numbers into a character string, and outputting the character string to a client for display. Data is automatically collected, the labor cost isreduced, and the operation and maintenance efficiency of the data center is improved.

The invention provides a method for batch extraction of similar target contours with single color in aerial images, and solves the problem of low batch extraction efficiency of similar target contoursrepeatedly appearing in aerial photos under the condition of similar target colors. The method comprises the following steps: (1) reading a small image where a target is located from an original image, and enhancing the image saturation difference; (2) converting a color space, generating a binary image, and carrying out contour extraction preprocessing on the binary image; and (3) extracting theobject with the maximum outer contour in the binary image as the contour of the target. The method has the advantages that the method is high in universality; the algorithm is simple; the calculationefficiency is high; result conciseness.

The invention provides a low-resolution image safety helmet recognition method and device. The method comprises the steps: carrying out edge feature extraction processing on two continuous image samples; comparing the processed two frames of image samples to identify an image change area; identifying a human body shape area in the change area of the next frame of image sample; carrying out characteristic analysis on a human body shape area to identify the safety helmet, utilizing an edge characteristic extraction technology and color and shape analysis to compare change areas of two continuousframes of images. The requirement for image quality is low, safety helmet identification can be carried out under the condition that the resolution ratio is very low, the requirement for system analysis hardware is lowered, and universality and adaptability are achieved.

The invention provides a real-time detection method for defect points of cloth. According to the real-time detection method, the real-time detection efficiency of defect points of cloth is improved when the accuracy is guaranteed, and particularly, the problems that a cloth pattern is difficult to be accurately positioned and then the pattern and the defect points cannot be accurately recognized by reasons of malposition movement due to mechanical sliding and the like after the cloth which is smaller than a standard size and is expected to be printed or jetted with a complex pattern is subjected to multiple printing or jetting processes are avoided. When the detection accuracy of the defect points which are 1mm or over higher than a peripheral region is guaranteed, the real-time detection of the cloth on a conveying belt in a real-time conveying period can be carried out, the detection speed is high, and the detection accuracy is high.

The invention relates to a transformer arc-shaped oil temperature meter reading method based on machine vision. The method comprises the following steps of S1, obtaining a dial image of a transformer arc-shaped oil temperature meter, and recording scales and pointers in the dial image, S2, determining the position of each pointer through color recognition, and determining a scale graph image according to the positions of the pointers, S3, determining the position of each scale through gray recognition, S4, according to the position of each pointer and the position of each scale, acquiring the scale closest to each pointer through a distance method, and S5, obtaining the temperature indicated by each pointer through the closest scale. According to the method, pointers and scales on a dial plate can be recognized through color recognition, the pointers correspond to the scales in position through a distance method, the relation between the scales in the arc-shaped oil temperature meter and the positions of the pointers are accurately read, and therefore the scale value of each pointer is accurately calculated and converted into corresponding temperature data.