25 results about "Thresholding" patented technology

The invention provides a vehicle information processing method, device and equipment and a computer readable storage medium. The method comprises the following steps: collecting a pavement image collected by an image collection device; identifying road information in the road surface image; identifying target vehicles in the road information, and determining center point information correspondingto each target vehicle; according to the center point information and a safe driving area, whether the target vehicle drives safely or not is judged, a judgment result is obtained, and the safe driving area is determined according to the center line of the road information and a preset range threshold value. Therefore, traffic factors such as road marker lines and traffic signal lamps in other roads do not need to be adopted for auxiliary calculation, the method can be suitable for traffic scenes of various roads, the recognition operation of whether the vehicle runs safely or not in the traffic scene of a single road is effectively limited, and the traffic safety of the traffic scene of the single road is guaranteed.

The invention relates to a cross-mirror pedestrian trajectory tracking method and system, an electronic device and a storage medium. An image collection device captures a portrait image, and forms first-level structural information according to the portrait image, a timestamp and the position information of the image collection device; a site server obtains the first-level structured information of the multiple image acquisition devices in the site range, obtains the portrait recognition confidence coefficient of the target pedestrian according to the first-level structured information, judges whether the portrait recognition confidence coefficient is larger than a face absolute confidence threshold value or not, and if the judgment result is yes, stores the first-level structured information in a target ID corresponding to the target pedestrian, and obtains second-level structured information from multiple pieces of first-level structured information of the target ID; and a cloud server obtains the second-level structured information of the plurality of site servers, and integrates the plurality of pieces of second-level structured information according to the timestamps and the position information of the image acquisition device to obtain the pedestrian real-time trajectory corresponding to the target ID, thereby improving the precision and real-time performance of pedestrian trajectory tracking.

The invention discloses a construction method and application of improved Focal Loss function for improving pavement disease segmentation effect. The method comprises the steps of setting a weight w of a Focal Loss function; presetting a threshold beta, and converting the weight w into a piecewise function w '; and optimizing the Focal Loss function by using the piecewise function w' to obtain animproved Focal Loss function. By means of the scheme, the method has the advantages of being accurate in classification, restraining interference caused by wrong labeling and the like, and has high practical value and promotional value in the technical field of image processing.

The invention discloses a face detection and recognition method. The method comprises the following steps: S1, preprocessing an image marked with a face frame to generate a training sample; S2, constructing a face detection and recognition network, wherein the face detection and recognition network adopts a deep learning network, and high-level features and low-level features of the network are fused; S3, inputting a training sample into the constructed face detection and recognition network for training until a training loss value is smaller than a preset threshold value, and completing training to obtain a deep learning network capable of outputting face detection and face recognition results. The face detection and recognition network is designed, face detection is regarded as a face center point problem, face center point detection and face feature vector extraction are combined for learning, a face feature vector corresponding to a face frame can be obtained while the face frame is obtained, the face feature vectors are compared to obtain a face recognition result, and a face detection result and a face recognition result are output through a network.

The invention discloses a regional backlight control method, a display and a storage medium. The method comprises the following steps: acquiring a regional gray feature value which corresponds to eachbacklight region and is used for representing gray information when a backlight module runs; comparing the regional gray feature value of each backlight region with a preset feature threshold value so as to divide each backlight region into a highlight region or a dark region; increasing the backlight brightness of all the highlight regions, and reducing the backlight brightness of all the dark areas. According to the invention, after the gray feature information of the backlight areas after affiliation division is extracted, the backlight of each backlight area can be adjusted according to the specific display content of the display and the required image representation, the image gray representation and the dynamic contrast effect are improved, and a more real and natural image is presented.

The invention discloses a non-contact heart rate variability feature extraction method based on a real application scene. The method comprises the following steps: 1) collecting an image containing a human face; 2) acquiring an image face area by using a face detection and tracking combined strategy; 3) enhancing skin color change through Euler amplification; 4) color space conversion and channel separation; 5) self-adaptive threshold skin detection; 6) extracting a source signal; (7) EEMD denoising; 8) carrying out five-point sliding smoothing filtering; 9) peak point detection and abnormal peak point correction; and 10) RR interval calculation and time domain, frequency domain and nonlinear HRV feature extraction. According to the method, the strategy for improving the feature extraction speed and the strategy for overcoming the influence of shaking and illumination and improving the feature extraction accuracy are integrated in the feature extraction method, the HRV features can be rapidly extracted in a non-contact mode in a real application environment, and it is guaranteed that the HRV features are consistent with contact type extraction results.

The invention discloses a vehicle locking method and device, a storage medium and electronic device. According to the method, image collecting equipment is installed on a vehicle body of a controlled vehicle, an environment image around the current position of the controlled vehicle is obtained through the image collecting equipment, and if the similarity between the environment image and a target image is larger than a first specified threshold value, the controlled vehicle is allowed to be in a locking state. According to the method, the vehicle lock is controlled, so that the vehicle can only be locked at the correct position, and the problem that the vehicle is parked at will is solved.

The invention discloses a self-supervised active learning method based on image classification. The method comprises the following steps: acquiring an image data set; setting the number of iterations and a threshold value; the self-supervised network carries out pre-training processing on the data to obtain feature mapping; the candidate unlabeled samples are evaluated according to the distance from the samples in the feature mapping to the center of the known category cluster in each round of iteration; querying a proper sample from an artificial expert; annotating a sample requested to be queried by an artificial expert, adding the annotated sample into an annotated pool, and updating an evaluation function and a cluster center; and inputting the sample into a classifier to train an optimization model, recording accuracy, and stopping iteration until the model reaches expected performance or a query sample exceeds a set upper limit. According to the method, the active learning indexes are constructed by fully utilizing feature mapping from self-supervised network learning to samples, and an active learning strategy is guided to perform sample selection, so that the labeling cost is saved.

The invention discloses a robust and efficient synthetic aperture radar multi-feature enhanced imaging method. The method comprises the following steps of: establishing an echo signal and a target multivariate feature enhancement framework; establishing a target multivariate feature enhancement solution equation, and solving target image data; applying entropy norm modeling synthetic aperture radar image focusing feature regularization prior, establishing a cost function, and solving an adjacent operator of the cost function to obtain a phase error and a focusing feature solution; applying l1norm modeling synthetic aperture radar image sparse feature regularization prior, and solving a sparse feature solution by using a soft threshold function and the phase error obtained in the step 3);and performing iteration on the target image data by using a Gaussian-Seidel method to obtain a synthetic aperture radar image with joint sparsity and focusing feature enhancement, and the like. According to the method, the sparse and focusing effects of the synthetic aperture radar image are recovered obviously in an experiment, and the focusing characteristics and sparse characteristics of the image can be enhanced quickly and effectively.

The invention is suitable for the technical field of image recognition, and provides a test tube recognition method and system, electronic equipment and a medium, and the method comprises the steps: obtaining the real-time image information of a target test tube rack, and recognizing a test tube rack region according to the real-time image information, a test tube being placed on the target test tube rack; acquiring equipment parameters of the target test tube rack, and extracting a test tube region according to the equipment parameters and the test tube rack region; acquiring position information of the shooting equipment, acquiring first depth data according to the position information and the test tube area, and comparing the first depth data with a first preset depth threshold to acquire a first depth comparison result; obtaining a target pixel number of the test tube area according to the first depth comparison result, and comparing the target pixel number with a preset pixel threshold to obtain a test tube identification result; by adopting the method, the problem that the test tube omission cannot be accurately identified in the prior art is solved.

The invention discloses an artificial intelligence translation integrated system, relates to the technical field of artificial intelligence translation, and solves the technical problem of low dialecttranslation accuracy caused by incapability of screening dialects of various regions in the prior art. A method comprises steps of analyzing dialect region data of the user through the region judgment unit, so that the region to which the dialect of the user belongs is judged; obtaining the keyword, obtaining the judgment coefficient Xio of the dialect region through a formula, judging that the corresponding dialect region is consistent if the judgment coefficient Xio of the dialect region is greater than or equal to a judgment coefficient threshold, and marking the dialect region as a pre-selected region; sorting the pre-selected regions according to a descending order of the judgment coefficients of the dialect regions, marking the dialect region corresponding to the judgment coefficient sorted first as a selected region, and obtaining a language library of the selected region through the Internet to convert real-time voice. The translation accuracy is improved, and the translationerror rate is reduced.

The invention relates to an unmanned aerial vehicle line patrol image auxiliary collection method and system based on resolution reconstruction, and the method comprises the steps: carrying out the image collection through an airborne camera of an unmanned aerial vehicle, inputting the collected image into an airborne computing device, and carrying out the following steps: carrying out the preprocessing of the image, and carrying out the collection of the image; evaluating the definition of the input image through a definition evaluation algorithm, and if the definition of the input image does not meet a set threshold value, judging that the input image is a low-definition image and inputting the low-definition image into a resolution reconstruction model; reconstructing the low-definition image through the resolution reconstruction model, and converting the low-definition image into a high-definition image; inputting the input image of which the definition meets a set threshold value and the high-definition image output by the resolution reconstruction model into a target detection model for inspection target detection; the airborne computing device is arranged on the unmanned aerial vehicle, resolution reconstruction and target detection are performed on the acquired image, the inspection target is detected on line, and the image quality is improved.

The invention discloses a slice processing method based on a complex background. The method comprises the following steps: converting a background image with uneven lighting through two-dimensional Fourier, and obtaining a frequency domain image; constructing a Gaussian low-pass filter, obtaining a picture after frequency filtering, and obtaining the frequency domain image; changing frequency of the Gaussian low-pass filter, cutting out different slice images after low-pass filtering, separating backgrounds and defects of the slice images, and completing slicing processing; for images of which the backgrounds and the defects of the slice images need to be further improved, carrying out scale stretching on filtered images after gray scale to obtain a frequency domain stretching image; performing threshold segmentation on the frequency domain stretching image, cutting out the slice images, completing background and defect separation of the slice images, and completing slice processing. The method has the beneficial effects that optimization is carried out from an algorithm, the cost of equipment is reduced, the development time period is shortened, and related defects and features are extracted from ato-be-detected object with a complex background through the algorithm, so that the method can be applied to different fields and is wide in applicability.

The invention provides an object three-dimensional shape reconstruction method and device, equipment and a storage medium, and relates to artificial intelligence image processing, and the method comprises the steps: carrying out the normalization of a two-dimensional image comprising a single object, and obtaining a normalized image; inputting the normalized image into an object shape reconstruction network, obtaining corresponding feature information through a feature extractor of the object shape reconstruction network, and inputting the feature information into a generator of the object shape reconstruction network to obtain a confidence three-dimensional tensor of a predicted voxel of the object; and carrying out binarization processing on each element in the confidence coefficient three-dimensional tensor based on a preset threshold value to obtain voxels representing the three-dimensional shape of the object. According to the method, the three-dimensional structure of the object can be accurately and efficiently recovered from the two-dimensional image, the efficiency of three-dimensional shape reconstruction is improved, and manpower and material resources are saved.

The invention provides an image detection method. The method comprises the following steps: carrying out image detection on at least one to-be-processed image by adopting a first model and a second model; when the second model performs image detection, determining an evaluation value corresponding to a first detection result in response to the first detection result obtained by the first model; and in response to a condition that an evaluation value corresponding to the first detection result is greater than or equal to a preset output threshold value, stopping detection of the second model on the at least one to-be-processed image, and outputting the first detection result. The invention further provides a processing core, electronic equipment and a computer readable medium.

The invention relates to the technical field of image processing, in particular to an image classification model construction method and device, an image classification method and device and electronic equipment, and the construction method comprises the steps of obtaining a first classification model and a second classification model, where the first classification model and the second classification model are obtained by training a sample image set, the complexity of the second classification model is greater than that of the first classification model, and the sample image set comprises sample images of a target category and sample images of other categories; connecting the first classification model with the second classification model by using a threshold judgment module to obtain an image classification model, where the threshold judgment module is used for determining whether the second classification model needs to be started based on the output of the first classification model. According to the invention, for the images which can achieve a better classification result by using the first classification model, the second classification model does not need to be started for classification, fusion based on lightweight and heavy models can be realized, the image classification time is greatly shortened, and high-efficiency image classification is realized.

The invention relates to a method for automatically extracting a coral reef based on a time sequence remote sensing image. The method comprises the following steps: 1, carrying out parallel preprocessing on the remote sensing image, namely carrying out atmospheric correction on the image; step 2, automatic screening of remote sensing images: automatic screening of the images is realized from four aspects of space overlapping, date uniqueness, cloud cover and image entropy; 3, time sequence construction: constructing a time sequence of the image MNDWI; and 4, automatically extracting the coral reef, namely constructing a characteristic curve of the coral reef time sequence, calculating a DTW value between the pixel-level time sequence and the characteristic curve, determining a DTW threshold value by using a dichotomy method, and extracting the coral reef. According to the invention, the problem that various noises exist in the coral reef image is solved, automatic screening of the remote sensing image is realized, a reliable method for automatically extracting the coral reef based on the time sequence remote sensing image is provided, and a process thought is provided for automatically extracting the coral reef range based on other satellite sensors.