158results about How to "Reduce detection error rate" patented technology

As disclosed herein, the present invention provides methods and systems for detecting, measuring, or monitoring the presence of absence of at least one emotion in a subject from a speech sample obtained from the subject. In particular, the methods and systems of the present invention comprise extracting at least one feature from the speech sample, assigning the speech sample a score using a weighted frequency band scoring scheme, and comparing the score with a general reference model or a control. As disclosed, the methods, systems and computer programs of the present invention provide detection error rates of about 4.7% or less, preferably about 3.0% or less, more preferably about 2.0% or less, most preferably about 1.0% or less and thereby reduced the detection error rate by about 60.0% or more, preferably about 70.0% or more, more preferably about 80.0% or more, even more preferably about 90% or more, and most preferably 95.0% or more than the detection error rates of prior art methods.

A digital watermarking apparatus that embeds two watermarks in host data is provided. The first watermarking block generates a plurality of watermarking location candidates of the first watermark, and embeds the first watermark into the respective location candidates of the host data, and then selects one candidate with a high level of robustness as the first watermarked host data. The second watermarking block scrambles the watermarking location information on the first watermark and thereby generates a plurality of the second watermark candidates. Then the second watermarking block embeds the respective second watermark candidates in the first watermarked host data and selects one candidate with a high level of robustness as the second watermarked host data.

The invention discloses a detection device and a detection method for series arc faults. The detection device comprises a continuous current collection module, which is used for continuously collecting current at a fixed sampling rate through an A/D (Analog to Digital) converter, a signal preprocessing module, which is used for carrying out moving average processing on collected data, an arc fault processing and analyzing module, which is used for computing a rate of change of current, a sum of previous N harmonics and a harmonic time domain waveform effective value, identifying an arc according to the three variables and giving a corresponding command, a sound-light alarm module, a drive module and a tripping control module, which are used for deciding whether sound-light alarm is carried out and a circuit is disconnected according to the command of the arc fault processing and analyzing module, and a communication module, which is communicated with the exterior through an RS485 interface. According to the detection device and the detection method, a time domain and a frequency domain are together utilized to judge whether the arc fault occurs, and through analyzing and testing, the method has the advantages of simplicity in operation, easiness in implementation, low cost and misjudgment rate and high detection probability.

The invention discloses a method for detecting a human body based on a range image, comprising the following steps: extracting an image characteristic according to the pixel of the collected range image; and inputting the image characteristic into a preset classification model to obtain whether the range image comprises the human body. With the method for detecting a human body based on a range image in the invention, the pixel of the range image is used for extracting the image characteristic so as to detect the human body and lower an error detection rate.

The invention provides a face recognition method and device. The method comprises the steps as follows: a to-be-recognized image containing a face is acquired; the to-be-recognized image is subjected to skin color segmentation on the basis of preset color space, and an area with the face skin color to be analyzed is determined in the to-be-recognized image; the area with the face skin color to be analyzed is subjected to face detection, and a face area is located; and the face area is subjected to face recognition, and a face recognition result is obtained. According to the method, the face recognition speed can be increased, and the face recognition reliability is improved.

Disclosed are an object detection method and an object detection device. The object detection method comprises a step of obtaining plural detection results of a current frame according to plural object detection methods; a step of setting initial probabilities of the plural detection results of the current frame; a step of calculating a movement frequency distribution diagram representing movement frequencies of respective pixels in the current frame; a step of obtaining detection results of a previous frame; a step of updating the probabilities of the plural detection results of the current frame; and a step of determining a final list of detected objects based on the updated probabilities of the plural detection results of the current frame.

The invention provides a system for automatically detecting electronic products through two-dimension codes and a method thereof. the system contains an information acquisition unit for acquiring detection information to be processed, a detection information processor which is connected with the information acquisition unit and is used for carrying out analytic processing on the received detection information to be processed and generating a corresponding two-dimension code of the processed detection information, a detection information display which is connected with the detection information processor and is used for displaying the two-dimension code of the processed detection information to a measured product to execute a corresponding action according to the processed detection information after acquisition of the measured product so as to determine a detection result of a corresponding detecting station where the measured product resides. According to the system and the method provided by the invention, repeated manual operations are not required. There is a fixed wired network connection between the detection information processor and an onsite monitoring and management system. Manual subjective judgment of a measured product is not depended. The detection result has reliability. Acquisition of information content is more and faster. Detection efficiency is raised, and labor cost is saved.

The invention discloses a method for detecting motion targets by cooperating multi-camera and solves the technical problems that the method for detecting motion targets by cooperating multi-camera of the prior art has high error rate. The invention utilizes multi-plane limited multi-camera cooperation to detect motion targets and solves the shelter problem by using multi-visual redundancy. Without increasing the number of cameras, the redundancy of multi-camera is efficiently removed by limiting multi-plane in order to remove false-alarm and decrease the error rate for detection. In the scene of three cameras and six persons, the average error rate of the prior art is decreased to 21.8% from 69.8%.

The invention discloses a mixed isolated island detection method. On the basis of mixing of artificial intelligence, voltage and frequency positive feedback of a decision tree (DT), the method based on artificial intelligence of the decision tree measures and stores characteristic quantity indexes of default events of selected target positions, information of a data collection is extracted and analyzed to train a DT classification device, and the operation state of DG (Distributed Generation) is judged to be whether network operation or isolated island operation through comparing the former data collection with a data collection in the case of pre-specified events. According to the method, on the basis of the DT method, the inverter control-based positive feedback method is further combined, complementarity and compatibility are strong; through introducing positive feedback disturbance, the characteristic quantity indexes can be obviously distinguished in the case of isolated island operation of the system and the network operation, judging the isolated island by the DT method is facilitated, and through increasing difference of the characteristic variable parameters in the case of isolated island operation or non-isolated island operation, the error rate of isolated island detection is thus reduced, and reliability of isolated island detection is improved.

The invention provides a pedestrian detection method based on a motion region. The pedestrian detection method comprises the steps of: acquiring video data photographed by video camera equipment; carrying out background modeling on the video data by using a Gaussian mixture model to obtain a motion foreground of a current frame; carrying out morphological processing on the motion foreground by applying an open operation, and eliminating isolated points surrounding the motion foreground; constructing rectangular bounding boxes enclosing the motion foreground, and deleting duplicate rectangular bounding boxes; acquiring coordinates, length and width of each rectangular bounding box; creating a motion region on an original image of the current frame according to the rectangular bounding boxes; and extracting HOG features from the motion region of the original image, and identifying the HOG features of the current frame by utilizing a pre-trained SVM classifier so as to determine whether a pedestrian is in the motion region.

The invention discloses a barrier detection method and a device thereof based on monocular vision. The method comprises the following steps: an image at the present moment is cut for obtaining a plurality of regions which are cut; self-vehicle motion parameters are obtained, and imaginary images at the present moment are generated by using all the pixels on the image t moment before the present moment after carrying out movement according to the self-vehicle motion parameters; the similarity level of the image and the imaginary image at the present moment is calculated, so as to determine candidate barrier area of the image at the present moment; whether each candidate barrier area has intersection with each cut area in the image at the present moment is calculated, if yes, the barrier judging percentage formed by the pixel area of intersection areas and the corresponding pixel area of the cutting areas is calculated, and barriers are determined according to the barrier judging percentage and preset barrier threshold. The method has the advantages of low faulty detecting ratio and complete barrier detection.

The invention discloses a method for extracting a closed edge image contour on the basis of an anisotropy Gaussian kernel. The method includes the steps that the anisotropy characteristics of edge directions are utilized, the multi-direction-angle anisotropy Gaussian kernel smooth filtering image is adopted, edge detection is then conducted, the structure of the edge contour is judged, the edges of fractures are filled or extend so that the edge fracture situation can not happen to the cross edges, and then the closed image edge contour is obtained. In addition, the method has the advantages of being high in noise steady performance, low in detection error rate, high in positioning accuracy and capable of achieving single-edge responding.

The present invention is suitable for the image identification field, and provides a human flow statistics method and a terminal device. The method comprises: determining a detection area in the imaging frame of a monitoring device; when detecting there is a detection target in the detection area, tracking the detection target to obtain the motion direction and the motion distance of the detection target; and when the motion direction and the motion distance of the detection target satisfy a preset condition, adding one to the statistics number of people. The detection range is limited in a small width area so as to improve the accuracy of target detection. The detection target is tracked to obtain the motion direction and the motion distance corresponding to the detection target, and when the motion direction and the motion distance satisfy the preset condition, it is determined that effective statistics number of people adds 1 so as to avoid interference of wandering personnel at a statistics area, eliminate an interference target from the range of the people statistics and reduce the error detection rate of the people statistics.

The invention provides a course error correcting method. The course error correcting method comprises the following steps: (a) collecting magnetic field measurement data, gyroscope measurement data and accelerometer measurement data; (b) calculating detection statistics; (c) determining whether a magnetic field is in a stable state or not in a corresponding time interval between an s step and a (s+m-1) step of a person; and (d) if the magnetic field is in the stable state, adopting a magnetic field solved course as an observation amount, estimating a course angle error by utilizing Kalman filtering, and correcting the course of the person by utilizing the course angle error. The invention also provides a course error correcting apparatus. The course error correcting apparatus comprises an accelerometer, a gyroscope, a magnetometer, a signal acquisition unit and a signal processing unit. The invention also provides a magnetic field detection method and apparatus.

The invention provides a building extraction method based on prior shape level set segmentation in order to overcome the shortcomings that the conventional building extraction technology is sensitive to noise and low in extraction precision, and depends on a segmentation result. According to the basic idea of the building extraction method based on the prior shape level set segmentation, urban building extraction is carried out in two steps, wherein in the first step, a rough outline of a building in altitude data is extracted, and morphological filtering and small-region removal are performed on the altitude data to obtain a final building distribution binary graph, and in the second step, outline information of the building is extracted in an optical image of an unmanned aerial vehicle, the building distribution binary graph serving as prior information is introduced into a level set segmentation model, and an evolution initial position and a prior outline are supplied to the model, so that the segmentation result is a building outline information extraction result. The building extraction method has the advantages of high pertinence and high precision.

The invention discloses a scene text detection method based on a long-term and short-term memory network. By using LSTM to model the feature, the invention obtains the spatial context relation of thefeature of the position and its upper, lower, left and right position, and then carries out text detection on the feature. Compared with the traditional method of using only local features, the scenetext detection method based on the long-term and short-term memory network provided by the invention can greatly reduce the error detection, can detect a lot of text information with inconspicuous local features, and reduces the omission of the text. The main innovation of the invention is to use LSTM to model the features horizontally and vertically, and to obtain the spatial context of the position and its upper, lower, left and right position features. Compared with the traditional scene text detection algorithm, this method can detect the text information with inconspicuous local featuresmore effectively, reduce the detection error rate and improve the detection accuracy.

The invention discloses a method and equipment for controlling information activation detection. The method comprises the following steps: determining a detection signal and a modulation constellation point of the detection symbol and control information; and according to the detection signal and the modulation constellation point of the detection symbol and the control information, judging whether control information is borne or not. According to the method and the equipment, the virtual detection probability of the control information can be effectively lowered.

The invention provides a method for classifying pixel blocks of JPEG images and image falsification detecting and falsified area positioning methods based on the same. The method comprises the following steps of: dividing to-be-detected JPEG images into continuous and non-overlapped 8*8 pixel blocks; taking one pixel block as the center and selecting a square area combined by (2n+1)*(2n+1) pixel blocks as a sub-image; utilizing a first number distribution probability feature of AC coefficients in the front i AC coefficient regions of the sub-image to divide the central pixel block into singlecompression or double compression; and confirming a classifying result as the double compression and a connected region with an area more than M 8*8 pixel blocks as a falsified area. According to themethod provided by the invention, the falsified area can be accurately detected and positioned; the falsified areas of three different falsifying technologies can be detected according to the method;the method is wide in application scope; the method is an automatic blind detecting method for JPEG falsified images and does not require any priori knowledge; and the detection coverage rate is higher and the error detection rate is low.

The invention relates to an echo canceling and phonetic testing method and apparatus for dialogue interactive front end, comprising following steps: using the telephone card containing recording and playing channel as voice interface card as the communicator dialogue interactive front end and using two channels of the telephone card to eliminate echo to get clear voice residual data; then processing the voice signal self-adapting detecting step to the voice residual data, judging the type of the voice residual data; and stopping echo elimination step and closing playing channel when judging the voice residual data as voice. The invention applies the recording and playing channels of the common telephone card, combining with voice enhancement, voice identification and signal processing means, to achieve the echo real-time elimination and voice real-time detection for dialogue interactive function.

The invention discloses a unilateral generalized gaussian model-based threshold method for SAR (Source Address Register) image change detection, belonging to the field of SAR image change detection. The threshold method comprises the implementation processes of: firstly, constructing a difference chart for two SAR images in the same region but in different times; solving a histogram of the difference images; then solving a probability distribution function of the histogram of an unchanged region by utilizing a unilateral generalized gaussian model; solving a probability distribution function of the histogram of a changed region by utilizing the gaussian model; determining a threshold automatically through a maximum posterior probability method; generating a change detection result chart through the threshold to complete the final detection of the changed regions in the two SAR images. In the invention, the fitting is carried out on a histogram curve of a threshold-generating region by utilizing a newly constructed model, therefore the precision of the final threshold is improved, and the threshold deviation caused by inaccurate curve fitting of the threshold-generating region is avoided to ensure that a better result is obtained through the SAR image change detection. Compared with several threshold methods for the SAR image change detection, the SAR image change detection result obtained in the invention has the optimal performance.

The invention relates to an aerial video moving object detection method based on multiple model estimation. The aerial photography video moving object detection method comprises the following steps: firstly, utilizing a Mean shift color segmentation method to segment a scene into a plurality of color blocks; then, utilizing dense pyramid luminous flux characteristics, and adopting a RANSAC (Random Sample Consensus) method to calculate an affine transformation model corresponding the color blocks with a large area; managing smaller color blocks, and analyzing point movement consistency in the blocks so as to carry out multiple model membership degree calculation, wherein the color blocks with the big membership degree are moving blocks, and otherwise, the color blocks are false alarm targets; and finally, combining and denoising the moving blocks, and outputting and displaying a detection result. The invention obtains a 5% detection error rate of a test in a public aerial video database, and the error rate is lowered by 5% than a traditional 10% error rate.

The invention relates to a distributed type grid-connecting system monitoring system, comprising a distributed type generation and grid-connecting interface device for monitoring the distributed type generation system and the load, uploading the monitoring information to a power dispatching agency and executing the remote commands from the power dispatching agency; a monitoring terminal for detecting the distributed type power supply grid-connecting points, the voltage, current and power values at the load points as well as the state of the grid-connecting switch and receiving the remote communication data of the distributed type power supply grid-connecting controller; and a power dispatching agency for realizing the operation state monitoring and control of the interfaced distribution type power supply distribution grids and realizing the energy management operation of the distributed type power supply according to its entrusted task scope. The invention further discloses an anti-island protection method for the distributed type generation and grid-connecting system monitoring system. According to the invention, without the introduction of a specially erected communication network, the cost for the system can be reduced. Under the premise that new distributed type power supply is added into the system, the monitoring system does not need more investment for its hardware. This meets the requirement for flexible interfacing of the distributed type power supply.

The invention provides an image steganography detection method based on Gabor filtering and a convolutional neural network, and belongs to the technical field of information hiding, and is characterized by comprising the following steps: selecting a carrier image and a secret-carrying image to generate a sample image; extracting steganography detection characteristics of the sample image; the steganalysis features and the class labels of the sample images are trained through an integrated classifier to obtain a steganalysis detector; and after steganography detection characteristics of the to-be-detected image are extracted, the steganography detection characteristics are input into the steganography detector for image steganography detection. Image filtering is performed by using a filterto construct a plurality of deep convolutional neural networks to carry out steganography detection feature learning, and the extraction of diversified learning type steganography detection characteristics is realized. Meanwhile, according to the method, the filtering coefficient is used for carrying out structural steganography detection feature extraction. Finally, the learning type steganography detection feature and the structural steganography detection feature are combined to serve as steganography detection features, steganography detection is carried out through an integrated classifier, and the steganography detection method remarkably reduces the detection error rate of image self-adaptive steganography.

The invention relates to an anti-collusion digital fingerprinting method based on a code division multiple access and diversity technology, comprising the following steps of: distributing a user identification code consisting of multiple stages of numbers for each purchaser; dividing an original carrier image into sub-images when fingerprints are embedded, then embedding all stages of numbers of a user into local DCT (Discrete Cosine Transformation) coefficients of the sub-images in all the sub-images by adopting a CDMA technology; and when finding a dubious copy, by combining with the original carrier image, firstly, detecting all stages of numbers and fingerprint strength values corresponding to the numbers of the dubious user in each sub-image, then synthesizing detection results of all the sub-images by adopting a diversity technology, and accurately judging one collusion member. The invention has less influence on the quality of the carrier image, low complexity of collusion detection and high detection precision rate.