77results about How to "Improve real-time detection" patented technology

The invention discloses a multi-sensor fusion-based autonomous obstacle avoidance unmanned aerial vehicle system and a control method. The system includes an environment information real-time detection module which carries out real-time detection on surrounding environment through adopting a multi-sensor fusion technology and transmits detected information to an obstacle data analysis processing module, the obstacle data analysis processing module which carries out environment structure sensing construction on the received information of the surrounding environment so as to determine an obstacle, and an obstacle avoidance decision-making module which determines an obstacle avoidance decision according to the output result of the obstacle data analysis processing module, so as to achieve obstacle avoidance of an unmanned aerial vehicle through the driving of power modules which is performed by a flight control system. According to the multi-sensor fusion-based autonomous obstacle avoidance unmanned aerial vehicle system and the control method of the invention, binocular machine vision systems are arranged around the body of the unmanned aerial vehicle, so that 3D space reconstruction can be realized; and an ultrasonic device and a millimeter wave radar in an advancing direction are used in cooperation, so that an obstacle avoidance method is more comprehensive. The system has the advantages of high real-time performance of obstacle detection, long visual detection distance and high resolution.

The invention discloses an image appearance based loop closure detecting method in monocular vision SLAM (simultaneous localization and mapping). The image appearance based loop closure detecting method includes acquiring images of the current scene by a monocular camera carried by a mobile robot during advancing, and extracting characteristics of bag of visual words of the images of the current scene; preprocessing the images by details of measuring similarities of the images according to inner products of image weight vectors and rejecting the current image highly similar to a previous history image; updating posterior probability in a loop closure hypothetical state by a Bayesian filter process to carry out loop closure detection so as to judge whether the current image is subjected to loop closure or not; and verifying loop closure detection results obtained in the previous step by an image reverse retrieval process. Further, in a process of establishing a visual dictionary, the quantity of clustering categories is regulated dynamically according to TSC (tightness and separation criterion) values which serve as an evaluation criterion for clustering results. Compared with the prior art, the loop closure detecting method has the advantages of high instantaneity and detection precision.

The invention discloses a mechanical arm safety collision strategy based on an external force observer. The mechanical arm safety collision strategy comprises the steps that firstly, a dynamic model of a flexible joint mechanical arm under external force interference is established and the stress is analyzed; a double encoder method is adopted to collect the angle information of the motor end andthe connecting rod end, and the external force observer needs to measure the joint position and speed and does not need to detect the acceleration q-diaeresis or an inverse matrix of an inertial matrix, so as to reflect the current input; and for a multi-degree of freedom mechanical arm, the computation of generalized momentum p and vector eta requires recursion or iteration methods, the observercan detect external torque or force, a dynamic threshold value is set according to the observation value of the observer, when the observed value exceeds the limit range of the dynamic threshold value, a control system considers that collision occurs, a touch stop mode is immediately switched to guarantee the safety of the operator and robots, and the safety responses at different motion speeds are achieved.

The invention discloses a double helix electrode capacitance tomography sensor for measuring an annular space that belongs to the field of an electrical process tomography device. The sensor comprises a main structure part and a capacitance measurement part. The main structure part is composed of concentric inner and outer tubes and a connection bracket used for fixing position. An annular space between the inner and outer tubes is a measured space. The capacitance measurement part comprises detection electrodes, electrode end shields, shielding cases, a signal transmission cable, ECT signal acquisition equipment and connection wires. By the utilization of electrode sensor parts which are respectively spirally arranged on the inner and outer tube walls, three-dimensional imaging of the measured annular space is truly realized. Due to the sensor structure with electrodes uniformly distributed on the two tube walls, imaging of the annular space is realized, and the problem of weak capacitance signal is also improved. The invention is a big breakthrough of the ECT technology.

The invention relates to a bridge vehicle wheel detection method based on a multilayer feature fusion neural network model. The method comprises the following steps of 1) constructing a deep learningneural network model based on multilayer feature fusion; 2) training the model by using the training sample data set; 3) carrying out image enhancement preprocessing operation on the to-be-detected image shot on the bridge; 4) inputting the preprocessed image into the model to obtain an output image with wheel vehicle category coordinate calibration; and 5) matching the detected wheels in the image with the corresponding vehicles by using an overlap rate measurement method. Compared with the prior art, the method has the advantages that the detection precision is improved, the real-time performance is enhanced, the probability of missed detection is reduced and the like, the wheel detection is realized while vehicle target detection is realized, and the vehicles on the same image are automatically matched with the wheels.

The invention provides a QRS-wave-group detection method. The QRS-wave-group detection method includes the steps that signals are pretreated to obtain an initial amplitude threshold value and an initial slope threshold value; the wave peak of a R wave is sought in the pretreated signals according to the initial amplitude threshold value and the initial slope threshold value, and the type and the end point of the R wave are judged according to the wave peak of the R wave; the type of a Q wave is judged according to the type and the wave peak of the R wave, and the wave trough of the Q wave is sought; the starting point and the end point of the Q wave and the wave trough of an S wave are sought, the type of the S wave is judged, and the starting point and the end point of the S wave are sought. The QRS-wave-group detection method is accurate in positioning and high in detection real-time performance, the starting and end points of special points can be positioned, more important diagnosis indexes are provided for electrocardiogram diagnosis, and the accuracy of the time limit of QRS waves can be increased.

The invention discloses a crop seedling and weed detection method and system based on deep learning, and the method comprises the steps: obtaining a crop seedling and associated weed image data set, and dividing the image data set into a training set, a verification set and a test set; labeling the training set image, the verification set image and the test set image, and amplifying the training set image; constructing an SSD detection model on a kernel deep learning framework, designing a lightweight dense connection network as a pre-feature extraction network, and fusing feature informationof different levels in the pre-feature extraction network and an extension network; inputting a training set into the improved SSD detection model for training to obtain a crop weed recognition and positioning model; inputting the test set into the trained model, and outputting types and position information of crops and weeds; according to the method, t. The problems of low detection precision, long consumed time and the like of a traditional crop and weed identification method are solved, and the accuracy and real-time performance of crop and weed detection are improved.

The invention discloses an intrusion detection system and method based on traceability information. The system includes a collector, a detector and an analyzer; the collector includes a traceability generation unit, a traceability trimming unit and a traceability storage unit, and the detector includes a rule base establishing unit , a rule matching unit and an output warning report unit, the analyzer includes a propagation query unit and a traceability query unit; the method is to collect the traceability information, store the traceability information in the form of a file on the file system, and store it in the traceability database at the same time. Based on the dependency information in the traceability database, a rule database is established based on the dependency information. During intrusion detection, the detected traceability information is compared with the traceability information in the rule base. When an intrusion is found, an early warning report is output, and the intrusion detection provided by the early warning report is used. The system and the method provided by the present invention improve the real-time performance of the intrusion detection.

The present invention discloses an outlier analysis-based device anomaly detection method and system. The method includes the following steps that: a plurality of monitoring data in a sewage treatment process are collected; the plurality of monitoring data are detected, so that outliers in the monitoring data and the distribution status of the outliers are obtained; and the outliers are screened and replaced according to the distribution status of the outliers; the anomaly of a device is determined according to the distribution status of the replaced outliers; and the abnormal status on the device is fed back. According to the outlier analysis-based device anomaly detection method and system of the invention, the anomalies of the monitoring data are sequentially analyzed, so that the outliers in the monitoring data can be obtained; and the distribution status of the outliers is further analyzed, so that the anomaly of the device can be detected, and therefore, detection accuracy and real-time performance can be improved.

The invention discloses a method for detecting a mobile phone playing behavior of a driver based on deep learning. On the basis of deep learning for target detection, a detection result is obtained; Preprocessing is carried out on an acquired video and a convolutional neural network of the video is optimized. A video sample of a mobile phone played by a driver in a cab is collected and converted into a frame image; The method has the advantages that the real-time performance and accuracy of deep learning detection targets are greatly improved, the calculation time is shortened to a great extent, and the feature extraction accuracy of hands and mobile phones is greatly improved.

The invention discloses a method and a system for detecting a rearview mirror viewing behavior of a driver. The method comprises the steps of acquiring a cab image by a vehicle-mounted video system; taking a face and neck visible skin region of the driver as an image analysis processing object; setting an algorithm for adaptive learning and tracing identification of the face and neck visible skin region; extracting a face and neck visible skin outer contour; taking a lowest point of the neck of the face and neck visible skin outer contour as a basic point; dividing the face and neck visible skin outer contour into a left part and a right part by a vertical line passing the basic point; defining an area ratio of the left and right parts as a feature parameter; obtaining a reference eigenvalue of the feature parameter; and judging whether the rearview mirror viewing behavior exists or not according to a numerical relationship between the current feature parameter and the reference eigenvalue in combination with a stare threshold and a duration threshold. The detection system is simple in structure; and the detection method has relatively high accuracy rate and good adaptive learning capability and anti-jamming capability, is high in detection efficiency, and can meet real-time detection and reminding requirements of the rearview mirror viewing behavior during quick running of a vehicle.

The invention discloses an unmanned aerial vehicle infrared target real-time detection method, which belongs to the field of unmanned aerial vehicle real-time monitoring, and comprises the following steps of: (1) constructing an IT-YOLO unmanned aerial vehicle infrared target detection network; (2) acquiring and manufacturing an infrared target detection data set of the unmanned aerial vehicle; (3) training to generate an unmanned aerial vehicle real-time infrared target detection model; and (4) detecting an infrared target by the unmanned aerial vehicle. On the basis of a Tiny-YOLOV3 lightweight target detection network, improvement of a trunk feature extraction network and replacement of a network detection layer are carried out; an IT-YOLO unmanned aerial vehicle infrared target detection network is constructed; an unmanned aerial vehicle infrared target detection data set is collected and manufactured under the vision of the unmanned aerial vehicle by using infrared imaging; and anunmanned aerial vehicle real-time infrared target detection model is trained and generated. The method is applied to an unmanned aerial vehicle to detect a target in a detection environment which cannot be dealt with by a visible light camera and a common digital night vision device such as a night lightless weather influence environment and a rainy and foggy weather influence environment, clearpedestrian and vehicle infrared spectral images can be collected at a distance of about 100 meters, and real-time and high-precision target detection is carried out.

The invention discloses a fatigue driving detection method integrating eye state and heart rate detection, which comprises the steps of S1, acquiring real-time video of a driver through a camera, and extracting a current frame to process; S2, performing preprocessing on a current frame image, and carrying out face detection by using a Adaboost algorithm; S3, performing positioning on the eyes on the basis of a detected face, and extracting an iris; S4, calculating a PERCLOS value; S5, installing an X-CTU which is communicated with equipment through the bluetooth, and recording the heart rate; and S6, performing fatigue judgment by integrating eye state and heart rate detection. The detection accuracy of the fatigue driving detection method under a simple background is close to 100% in the face detection process; a classic PERCLOS algorithm is improved, and the function can be realized through coarse positioning for an eye region without a need of knowing the exact position of the eyes or calculating the closure degree of the eyes; two parameters of eye state and heart rate detection are integrated, and both the detection accuracy and the real-time performance are improved.

The invention discloses an operating vehicle-based rail nucleus flaw detection system and method. The system comprises an acceleration sensor, a signal acquisition and preprocessing unit and a signal processing unit; the acceleration sensor is used for generating horizontal and vertical acceleration signals of a steering frame according to vibration generated in a vehicle running process; the signal acquisition and preprocessing unit is used for acquiring the horizontal and vertical acceleration signals with a fixed sampling frequency and preprocessing the signals to generate signals to be detected at all sampling times; the signal processing unit is used for sequentially performing time domain characteristic value analysis for identifying rail nucleus flaws, continuous wavelet analysis for identifying the rail nucleus flaws and empirical mode decomposition for identifying the rail nucleus flaws on the signals to be detected at the all sampling times so as to finally determine a to-be-detected in which the rail nucleus flaws exist. According to the technical scheme of the system, real-time detection of rails when the vehicle passes the rails can be realized without time limit, and a problem of high human error possibility during manual detection is avoided.

The embodiment of the invention discloses an object detection method and related equipment, and the method comprises the steps: obtaining a to-be-detected depth image, and segmenting a foreground region from the depth image according to the depth values of pixel points in the depth image; because the target object has a foreground area which is more likely to be in the depth image, the corresponding point cloud can be generated for the foreground area. Therefore, whether the depth image includes the target object is detected according to the generated point cloud. According to the method, thepoint cloud is generated from the foreground area which may more possibly include the target object in the depth image and is further detected, and the background area which may not include the targetobject is eliminated, that is, the point cloud is not generated from the areas, so that the calculation amount is reduced, and the object detection efficiency and the detection real-time performanceare improved.

The invention relates to the technical field of radar target detection, and in particular, relates to an FPGA-based two-dimensional ordered statistical constant false alarm detector implementation method. According to the method disclosed by the invention, detection is carried out in each clock period from two-dimensional power matrix input, so that the repeated reading time delay of the data cache is reduced, and the detection speed can be effectively improved. Aiming at the condition of large calculated amount in a traditional OS-CFAR (Ordered Statistical Constant False Alarm Rate) sorting process, the FPGA implementation method is completed by adopting parallel comparison, so that the sorting complexity is reduced. The reusability of a module is fully considered, two-dimensional slidingwindow structure setting is completed through a shifting register set, two-dimensional OS-CFAR detection is achieved, and BRAM memory resources and DSP resources of the FPGA can be saved.

The invention provides a building security system and a building security method based on pulse light spot array pattern change detection. The building security system comprises a camera, a mainframe, a warning device and a light source, the light source gives out dot matrix rays by a pulse irradiation method, the dot matrix rays irradiate on a detection area to form a light spot array, a sensor for detecting the light spot array is arranged in the camera, and a grating used for adjusting shape of the light spot array and number of light spots is further arranged in front of the light source. Real-timeness and improved efficiency of detection can be achieved by adjusting the light spots and parameters of an image sensor, the building security is easy to implement and low in cost, and has a great application prospect. The building security method is capable of realizing intrusion detection within a variable range by changing the number of the light spots, light spot array pattern and position of the light source for changing detection range.

The application provides a method and apparatus for detecting the hang of an FPGA chip, the method comprising: obtaining, within a preset time period, a packet receiving statistical value and a packet transmitting statistical value for each port on a switch chip and a packet loss statistical value of a port on the FPGA chip connected to the port, and determining whether the packet transmitting statistical value is not zero and the sum of the packet receiving statistical value and the packet loss statistical value is less than a first predetermined value, and if so, adding 1 to an error number; determining whether the error number is higher than a second preset value; if so, determining the hang of the FPGA chip. The method and apparatus determine whether the port is faulted according to the packet receiving statistical value, packet transmitting statistical value and the packet loss statistical value, and can determine whether the FPGA chip is subjected to hang according to the number of error ports, prevents manual detection, improves real-timeness of the detection.

The invention provides a detection method for the concentration of low-concentration NO and SO2 mixed gas. By utilizing the characteristic that the spectral absorption interference between SO2 and NO gas at the 206-212nm wave band is low, through data extraction, spectral absorption data of the SO2 gas in the 206-212nm wave band is separated, and the concentration of the SO2 gas is determined at first; further, by utilizing a spectral superposition principle of the mixed gas and spectral absorption data of a 200-230nm wave band range where the spectral absorption of the SO2 and NO gas is seriously mixed and superposed, the spectral absorption data at the 200-230nm wave band is analyzed out, and therefore the concentration of the NO gas is determined, so that the respective concentration detection of the SO2 gas and the NO gas in the NO and SO2 mixed gas is achieved; the method can have good detection accuracy, and the problem is solved that concentration detection methods in the prior art cannot meet the requirement of detection precision regarding the low-concentration No and SO2 mixed gas at 100 ppm or below; meanwhile the amount of data involved in the whole detecting process is small, and therefore the method is conducive to speeding up the calculation of detection processing and improving the real-time performance of the detection.

The invention discloses a method for detecting the state of a ground wire in a substation. A ground wire controller is installed on a ground connection wire; a ground connection pile coding and a ground wire cabinet coding are respectively installed on a ground connection pile and a ground wire cabinet; when the ground connection wire is connected to the ground connection pile, the ground wire controller can detect the ground connection pile coding of the ground connection pile; when the ground connection wire is placed in the ground wire cabinet, the ground wire controller can detect the ground wire cabinet coding of the ground wire cabinet; when the ground connection wire is for standby, the ground wire controller can not detect the ground connection pile coding and the ground wire cabinet coding, transmits the coding information to a base station through a wireless network, forwards the coding information to a host through the base station, judges the state of the ground wire by thehost, and can conveniently realize complete detection of three states, such as assembling and connection, standby and storage of the ground connection wire.

The invention discloses a phishing application detection method. According to the method, view information of a current application interface is acquired and analyzed, and the category of the currentinterface is judged according to preset rules; when the category of the current interface is a preset category, text information relevant to an input box and text information relevant to a clickable view are acquired respectively; content needing to be input into the input box is judged according to the text information relevant to the input box, and then preset taint data of the category of the interface is acquired to complete data input; and clicking operation is performed according to the text information relevant to the clickable view and the preset rules to trigger a corresponding behavior. The invention furthermore discloses a phishing application detection device. The device comprises a taint data module, an interface analysis module, an analog input module and an analog clicking module. The scheme has the advantages of being high in detection instantaneity, convenient to realize, high in detection efficiency and the like and is also suitable for simultaneous detection of large-scale phishing samples.

The invention provides a medium and long wave infrared image fusion method and device applied to the ship detection field under a sea-sky background and relates to the field of ship detection technology. The method comprises the steps that a first image and a second image are acquired, wherein the first image and the second image are a medium wave infrared image and a long wave infrared image of ato-be-detected scene respectively; wavelet-Contourlet transformation is performed on the first image and the second image to obtain two image sets, wherein each image set comprises multi-scale multi-direction four-layer decomposition images of the first image or the second image; the decomposition images on the same layer of the two image sets are fused by use of the PCNN global coupling propertyand the synchronous pulse emission characteristic in combination with a structural similarity operator, and four fusion image layers are obtained; and the four fusion image layers are reconstructed to obtain a photographed image of the to-be-detected scene. Through fusion of the medium wave infrared image and the long wave infrared image, the instantaneity and motility of capturing the to-be-detected scene are enhanced, and the technical problem that an existing ship detection technology is poor in instantaneity and motility is relieved.

The invention provides a traffic sign detection method for unmanned drive. The method comprises the steps of performing multi-scaling on an image, generating integral images, selecting feature images,and sliding to traverse the images, wherein the multi-scaling is performed on the input image to be detected, and thus multiple scaled images are generated; the feature images, comprising color information and gradient information, of the scaled images are computed, and the integral images of the feature images are computed; features of the integral images are selected via a feature selection classifier, and thus selected feature images are generated; and the selected feature images are traversed via a sliding window, and traffic signs in the sliding window are detected. The invention relatesto electronic equipment and a readable storage medium, used for executing the traffic sign detection method for unmanned drive; the invention also relates to a traffic sign detection system for unmanned drive; according to the method, equipment, medium and system provided by the invention, the image processing efficiency is improved, the traffic sign detection is high real-time and the detectionaccuracy is high.

The invention discloses an active hiding method when a patrol robot executes a tracking task. The method comprises the following steps: (1) suspicious target detection: detecting a suspicious target by using an HOG feature descriptor and an SVM support vector machine, and tracking the detected suspicious target by using a KCF; and (2) hidden landform perception: utilizing a Unet semantic segmentation network to segment a hidden landform in a complex environment in real time; and (3) behavior decision-making: carrying out fusion processing on hidden landform information and suspicious target information, and deciding a tracking action of the robot according to a result. Through the active hiding method, the patrol robot can sense the hidden landform in the complex environment and the safety state in the tracking process, the hiding method is planned, the probability that the patrol robot is found and damaged is reduced, and the invisibility of the robot during patrol is enhanced. The invention solves the problems of task failure and economic property loss caused by lack of active self-protection measures of patrol robots in the prior arts.