119results about How to "Run in real time" patented technology

The invention discloses a simultaneous positioning and map building method which can reliably handle strong rotation and fast movement. The method proposes a simultaneous positioning and map building algorithm framework based on key frames, which can support fast local map expansion. In this framework, a new feature tracking method based on multiple homography matrix is proposed, which is efficient and robust under strong rotation and fast motion. A camera orientation optimization framework based on a sliding window is also proposed, and motion constraints are added between successive frames with simulated or actual IMU data. Finally, a method of obtaining the real scale of a specific plane and scene is proposed, so that a virtual object is placed on the specific plane in real size.

The invention relates to a mobile robot real-time layered path planning method based on a grid map. The method, which is used in a grid map, comprises the following steps: carrying out binary classification according to external environment information, and establishing a grid map; planning an outer path according to a speed-based artificial potential field method; and when the path is about to fall into local minimum and oscillation, planning an inner path according to an A star algorithm. The problem that the traditional artificial potential field method may fall into local minimum and planned path oscillation and the problem that the A star algorithm is not applicable to real-time path planning are solved. Real-time, safe and stable operation of mobile robots is ensured.

The invention relates to a finger clicking character recognition method and a translation method based on artificial intelligence. The recognition method comprises the following steps of (1) respectively constructing and training each neural network; (2) using the acquisition device to acquire the current image of the pre-detection area in real time, and continuously inputting the image into the pre-trained finger positioning neural network to obtain the finger position information under the finger click state; 3) taking that position of the user's fin as the center, intercepting the image ofthe box area, inputting the angle recognition neural network, and outputting the rotation angle of the text in the image area; 4) rotating the rotation angle to intercept the frame area image with theposition of the user's fin as the center, and outputting the position information and the size information of the detected character area; (5) intercepting a corresponding image, inputting the OCR recognition neural network, and outputting the recognized text content. The invention not only improves the identification efficiency but also enables the identification artificial intelligence to be realized.

The invention relates to a real time dynamic rescheduling method used for a process industry production scheduling system, aiming at solving the problems of present domestic process industry enterprises that production planning and dispatching have low efficiency, poor accuracy and real-time capability. The real time monitoring method of the invention includes the establishment of a scheduling-optimization math model and the working-out of a dispatching-optimization plan, the completion of the execution of production dispatching instructions and timely tracking calculation; a rolling optimization method includes the arrangement of dispatching time periods, the calculation of the state changes of various production data from present time to scheduling starting time, the generation of the data files of a target model and the establishment of the scheduling-optimization math model. The method of the invention is characterized by real-time, flexibility and promptness, etc. and can realize the integration of the control, the operation, the planning and the management of production-process and enterprise management to realize the balanced and high-effective production of enterprise logistics, production capability and source, thus enhancing the integral production efficiency of enterprises and reducing the cost of production and stock.

The invention discloses a simultaneous positioning and dense three-dimensional reconstruction method, and the method can process quick motion and frequent closing of a loop in a robust manner, and can operate all the time in a larger-scale scene. The method not only can be used for scanning a high-quality three-dimensional model, but also can meet the demands of virtual reality and augmented reality. The method proposes a simultaneous positioning and map construction framework based on a key frame. At first, the method employs the depth and color information at the same time, and can operate on a CPU at a high speed based on the positioning of the key frame, and also can operate in a challenging scene in a robust manner. In order to reduce an accumulated error, the method introduces increment bundling adjustment, can greatly reduce the calculation amount, and enables the local and global bundling adjustment to be completed in a unified framework. Secondly, the invention proposes a fusion method based on the key frame, and the method can achieve the online generation of a model and updates the three-dimensional model in real time during the adjustment of the posture of the key frame.

The invention discloses a video geographic information system and method for urban areas. The system comprises a video acquisition Agent, a video server and a geographic information system (GIS) positioned in the server, wherein the Agent is composed of a video transmission module and a video image projection transformation module; and the video server is composed of a video image and map registering module, a video projection and geographic information integrating module and a dynamic video GIS display module. The system of the invention is independent, data of the system comes from video monitoring cameras in the urban area, the instantaneity is high, lots of cameras can be simultaneously connected to the system, a lot of continuous monitoring data can be simultaneously read from one map, and visual whole-city running states and a target continuously searching and tracking service are provided for the city managers and decision makers.

The invention discloses an unmanned aerial vehicle attitude control system and method based on self-adaption complementation fusion. The method is characterized by, to begin with, reading sensor data and carrying out corresponding filtering on the data; then, carrying out self-adaption complementation fusion according to frequency-domain characteristics of different sensors to obtain aerial attitude information of a current aerial vehicle, and furthermore, obtaining an Euler angle needed by control; next, carrying mathematical modeling on the aerial vehicle and designing two controllers, wherein one controller is used for controlling the pitch angle and the roll angle, and the other controller is used for controlling the course angle; and finally, carrying out superposition on the outputs of the two controller to obtain a total output, and realizing stable control of the aerial vehicle by changing the rotation speed of a motor through a driver. The system and method realize attitude wide-range high-precision calculation, quicker data convergence and higher precision.

The invention relates to an attention mechanism CNN-based 5-day and 9-day incubated egg embryo image classification method so as to realize the egg embryo activity detection and sorting. The method comprises the following steps of 1) proposing to use the depth separable convolution to realize the more sufficient feature extraction on an existing channel attention feature map; 2) generating a high-resolution and large-receptive field attention weight map by adopting the spatial pyramid cavity convolution; and 3) performing the spatial weighting on the feature map by the weight mask through element-by-element multiplication, so that the effect of enhancing the useful information and suppressing the noise can be realized. Results show that the attention module plays a role of a feature selector in the convolutional neural network, and the feature expression capability of the convolutional neural network is enhanced, so that the classification accuracy is improved, and the problems of lowefficiency, high labor consumption and the like of manually sorting the 5-day and 9-day incubated egg embryos are successfully solved.

The invention provides a wake-up signal detection method for an underwater acoustic communication MODEM. The method includes the steps that two repeated two-phase code signals generated by a balanced Gold sequence are selected as wake-up signals, sliding correlation treatment between adjacent data block pairs different in length is adopted in a receiving end, a data block start moment corresponding to the maximum correlation value is wake-up signal arrival time, the difference between the length of a data block corresponding to the maximum correlation value and the length of each sent wake-up signal is the Doppler time stretching length, and therefore the wake-up signal arrival time and the Doppler time stretching length are jointly estimated; a current correlation value is obtained in the mode that a product of start data of the currently adjacent data block pairs at last time is subtracted from the sum of a former correlation value and a product of tail data of the currently adjacent data block pairs, the calculation amount is far smaller than that of multiple sliding copy correlation devices, and the method can be conducted in real time under the conditions of dormancy of the underwater acoustic communication MODEM and the low clock frequency of a processor and is high in practicality.

The invention relates to a real-time online multi-target tracking method adopting a temporal dynamic appearance model. The method comprises the following steps: taking appearance information of each target in a multi-target tracking process as an appearance sequence of temporal orderliness; expressing the temporal dynamic variation characteristics of the appearance sequence by establishing an appearance model with strong discriminating ability by considering the rule of change of the target appearance information in the tracking process following the change of time; supplying accurate appearance matching information related to the data between target and detection results through structural diversity of the modeling of hidden state observation models in a characteristic space; relating the detection results with the target frame by frame, thereby finishing the real-time online multi-target tracking task and accurately real-time tracking a plurality of interesting targets in a tracking video. The tracking property is ensured while calculation efficiency is ultra-high. The real-time online multi-target tracking method can be practically applied to various real-time systems.

According to the technical scheme of the invention, a virtual reality module and a control module are in real-time communication with each other. After the related personnel compile a program through the control module, the control module outputs a real electric signal. The electric signal is converted into a virtual control signal through a communication interface module, and then the virtual control signal is transmitted to the virtual reality module in real time. The virtual reality module controls a production line to conduct operations in real time according to the virtual control signal. At this time, the related personnel can check whether equipment conduct operations according to the program or not in the operation checking manner through the live-action observation in a virtual reality environment. Meanwhile, during the operation of the virtual equipment, if a virtual sensor is triggered, a virtual reality system sends a virtual feedback signal to the communication interface module. The communication interface module converts the virtual feedback signal into a real feedback signal and then sends the real feedback signal to the control module. Therefore, the closed-loop online real-time simulation is realized, while no actual production equipment (such as motors, hydraulic systems, various electromechanical devices, etc) needs to be installed. As a result, the teaching purpose and the experiment simulation purpose are realized.

The invention relates to a background modeling acceleration method which comprises the following steps of: filtering an image imported into a CUDA (Compute Unified Device Architecture); establishing Gaussian models for corresponding pixels by each thread, wherein the parameter data of the Gaussian model required by a current thread block is written into a shared memory, and after calculation is finished, a calculation result is written into a global memory from the shared memory. The method has universal applicability, and no matter for a scene with a large traffic flow or a scene with a small traffic flow, consistent acceleration can be obtained basically. The method is applied to the field of intelligent video monitoring for improving the efficiency of a background modeling algorithm so as to ensure that a system can guarantee to be run in real time while facing a larger data volume; and moreover, superfluous resources can be assigned to other algorithms, which provides a certain space for improving other functions. The method is applied to object clustering in the field of computer vision for accelerating a clustering process of objects and reducing a large quantity of waiting time.

The invention relates to a new method for generating a vehicle-mounted driving assisting image. The new method aims at solving the problem that in a vehicle-mounted look-around system, the imaging effects of images of nearby views, distant views and three-dimensional objects on a virtual camera can not respectively have the perspective effect truth. According to the new method, it is proposed that the mapping relation between a practical camera and the virtual camera is built through a spherical virtual projective plane, several virtual camera placement modes are proposed according to different vehicle driving states, the burden that coordinate transformation needs to be carried out when a driver views images of the practical camera can be reduced through the virtual camera placement modes, and easily-identified vehicle surrounding images are also provided for the driver. In order to quicken the processing speed, a forward mapping table is proposed in the new method: firstly, a mapping table for mapping the images of the practical camera to the spherical virtual projective plane is built, then the mapping table is mapped to the spherical virtual projective plane again according to the position of the virtual camera, and therefore the forward mapping table for mapping the images of the practical camera to the images of the virtual camera is obtained.

The present invention provides a monocular SLAM method that can create large-scale maps by using direct image alignment correction, explicit absorption and detection of scale drift; then adding filter-based semi-dense depth map estimation to achieve continuous With large-scale environment maps, the method not only tracks camera motion locally, but also maintains and tracks it on a global map of the environment. At the same time, the method can also run in real time on the CPU of an ordinary PC, and as a ranging method, it can even run on modern smartphones.

The invention relates to a navigation and position equipment in the electronic technical field and real-time processing thereof, in particular to a real-time software receiver of the global position system, which implements real-time location by using the received GPS signal from the global position satellite system and is used for the signal sampling and the computer software processing, as well as a real-time processing method of the invention. The apparatus consists of a complex programmable logic device and an interface module of serial-to-parallel conversion, etc. The method executes the processing of baseband signal for the computer software under the windows XP environment, the method mainly resolves the related technical problems of how to implement the interfaces of signal sampling and transmission modules and the processing of baseband signal, etc. The invention has the advantages that the apparatus implements the real-time execution in the windows XP system of Microsoft, and the software adopts the bit arithmetic way, which reduces the computation, accelerates the processing speed and achieves real-time location, and higher precision and sensitivity through the rapidity of signal processing.

The invention relates to an embedded type real-time scheduling control device and method based on an MCU. The device comprises a microprocessor, a timer, a storage, input equipment and output equipment. The input equipment comprises an I / O interface, an analog-digital converter module and a communication interface. The storage obtains data and instructions transmitted by a sensor and a bus in a communication mode through the input equipment. The software instructions and the data are stored in the storage. The microprocessor takes out the software instructions and the data in the storage according to time base signals and software scheduling control rules provided by the timer to carry out operation, processing and judging, and processing results are obtained and output to the output equipment. The output equipment comprises an I / O interface, a pulse width modulation module and a communication interface. Output executing results are transmitted through the output equipment in different signal modes. The embedded type real-time scheduling control device and method have the advantages that the independence of continuous tasks and discrete tasks is guaranteed, and meanwhile the operation efficiency and the real-time performance of the tasks are improved.

The invention relates to the technical field of automatic driving, in particular to an automatic parking space detection method based on deep learning. The method comprises the steps of summing the loss function obtained on the basis of deep learning and the cost function of the template obtained on the basis of template matching to obtain a total cost function, and searching the total cost function by taking the position of the point with the maximum probability of any model as an initial value to obtain the position of the parking space. Compared with the prior art, the method has the advantages that the method is easy to implement; according to the method, template matching and deep learning are combined to carry out parking space detection in the image; the method is effective under the conditions of good illumination conditions and good splicing effects, and can realize robust and accurate parking space detection under the conditions of complex light sources, road surface reflection, interference in parking spaces, inaccurate splicing top views and the like. And the calculation complexity of distance transformation and Canny edge detection is very low, the calculation amount is very small, and the method can be operated on an embedded platform in real time, so that the method has a direct application value.

In various examples, sensor data may be received that represents a field of view of a sensor of a vehicle located in a physical environment. The sensor data may be applied to a machine learning modelthat computes both a set of boundary points that correspond to a boundary dividing drivable free-space from non-drivable space in the physical environment and class labels for boundary points of the set of boundary points that correspond to the boundary. Locations within the physical environment may be determined from the set of boundary points represented by the sensor data, and the vehicle may be controlled through the physical environment within the drivable free-space using the locations and the class labels.

The invention belongs to the video monitoring field and provides a video based motion pattern detection method and device. The video based motion pattern detection method comprises obtaining a video image; estimating a motion foreground image, wherein the estimation of the motion foreground image specifically comprises establishing a preliminary motion foreground image, extracting a contour image of the motion foreground image and establishing a motion history image for the contour image; detecting a moving target of the motion foreground image; updating the state of the moving target. The video based motion pattern detection device comprises an image acquisition module, a motion foreground estimation module connected with the image acquisition module, a moving target detection module connected with the motion foreground estimation module and a target state update module connected with the moving target detection module. According to the video based motion pattern detection method and device, the problems of the slow background update caused by background modeling and the serious prospect cavity are solved, the accurate and stable continuous counting on the movement displacement and movement direction can be achieved, and the effective accurate information can be provided for application development.

The invention discloses a ground clutter recognition method, which comprises the following steps of: detecting a signal-noise ratio of a radar echo signal and judging whether the signal-noise ratio is less than 3 dB; if the signal-noise ratio is less than 3 dB, calculating three characteristic quantities TDBZ, SPIN and CPA (Clutter Phase Alignment) of the radar echo signal, wherein nine range bins with a current range bin as center are selected by TDBZ and 11 range bins with the current range bin as center are selected by SPIN; converting each characteristic quantity into a value of 0-1 through a respective membership function respectively; calculating the three characteristic quantities with values of 0-1 by applying fuzzy logic to obtain ground clutter probability; and judging the ground clutter probability by using a certain threshold value to obtain a ground clutter recognition result. According to the technical scheme provided by the embodiment of the invention, the ground clutter can be efficiently recognized, super refraction is adaptively treated, the calculation is low, real-time operation at a radar signal processing front end can be realized, and the loss of a weather signal can be avoided.

The invention relates to a prostate transrectal ultrasound (TRUS) image segmentation method based on a convolutional neural network. The method comprises the following steps: 1) extracting multi-scaleconvolution features in a prostate TRUS image by using an extended spatial pyramid pooling (DSPP) module with global feature coding semantic information; 2) in order to process the missing boundary in the segmented prostate shadow region, proposing a super-hybrid feature (SHF) capable of mixing low-level features and high-level features; and 3) a novel network composed of an encoder and a decoderis provided. The result shows that the method can achieve the precise segmentation of the prostate TRUS image, improves the segmentation accuracy and speed compared with the prior art, and successfully solves problems that the manual segmentation efficiency of the prostate TRUS image is low, the accuracy is low, and excessive manpower resources are consumed.

The invention provides a multi-point touch tracking identification method and a multi-point touch tracking identification system. The method comprises the following steps of: acquiring touch point positions in two adjacent frames of images respectively; calculating the distance between the touch points of the images of the front frame and the rear frame according to the touch point positions; screening actual association between the touch points according to the distance, and gradually searching the association modes of the touch points of the images of the front frame and the rear frame according to the sequence of the distance of the touch points; and calculating the touch point dispersion degree of the association mode searched every time, and judging that the previous association modeis an optimal association mode when the touch point dispersion degree of the later association mode is greater than the touch point dispersion degree of the previous association mode for the first time. According to the multi-point touch tracking identification method and the multi-point touch tracking identification system, accuracy and robustness of the association of the touch points in the front frame and the rear frame are ensured by global analysis. By screening the actual association between the touch points, the tracking identification efficiency is improved.

The invention provides a real-time large-scene three-dimensional semantic modeling method. The method comprises the following steps of S1, constructing a three-dimensional geometric model through an RGB image and a depth image obtained by scanning a scene through a sensor; s2, inputting the three-dimensional geometric model into a three-dimensional convolutional neural network to complete semanticsegmentation; s3, integrating semantic tags output by the three-dimensional convolutional neural network into the three-dimensional geometric model to complete semantic modeling; wherein the step ofconstructing the three-dimensional geometric model and the step of semantic segmentation are combined in a multi-thread mode and are carried out at the same time. Joint real-time three-dimensional geometric reconstruction and semantic reconstruction are realized; by adopting the sparse convolutional neural network and accelerating the calculation of the convolutional network, the performance of real-time operation can be achieved; according to the method, two-dimensional convolution in UNet is replaced by three-dimensional convolution, and meanwhile, the depth of the UNet network is increased,so that the capacity of the convolutional network is larger, and the semantic segmentation accuracy is improved.

The invention discloses a method for restoring a moving track of a hand in the air by using a radio signal, which is free from a special sensor and is used for detecting and restoring a writing trackof a finger in the air by using the radio signal of existing communication equipment. One signal transmitting device Tx and two signal receiving devices Rx are placed, each receiving device is provided with at least two antennas, and no barrier exists between the Tx and the Rx. The hand does not need to wear any equipment. The moving area of the hand in air is interior of a square area defined byTx-Rx1 and Tx-Rx2. A radio signal is disturbed by using the movement of the hand, wireless channel changes on a plurality of receiving antennas are measured and extracted, and then the movement trackof the hand in the air is calculated. According to the method, the channel CSI value acquired by the existing wireless communication equipment is utilized, and extra hardware equipment or hardware change is not needed, so that the cost is saved. The application range is wide. Accurate tracking can be carried out, and the track restoration precision is high. Real-time continuous operation can be achieved, and convenience and practicability are achieved.

The present invention discloses a traffic intersection video and computer parallel processing-based real-time pedestrian early-warning method. The method comprises the steps of extracting the moving foreground of an intersection, extracting pedestrian targets in the classified manner, tracking pedestrian targets in real time and alarming the condition of a pedestrian entering the intersection. During the early-warning process, moving targets in the foreground of a monitoring video is extracted based on the Vibe algorithm. After that, pedestrian targets and non-pedestrian targets, out of all moving targets in the foreground, are classified by using an off-line trained pedestrian linear SVM classification model. Finally, pedestrian targets are tracked based on the joint probabilistic data association (JPDA) tracking algorithm, so that the moving data of pedestrians are acquired. Therefore, the early-warning is conducted. According to the technical scheme of the invention, a traffic intersection video is detected, and a pedestrian target in the traffic intersection video is tracked and processed in real time. In this way, an improved method is provided. The multi-thread parallel processing is conducted and the shared data of queues are buffered dynamically. As a result, on the premise that the calculated amount of the foreground detecting and tracking algorithm is relatively large, the monitoring video can be maximally ensured to be processed in real time.

The invention relates to a modeling method and system for calculating energy efficiency of motor driven fluid transportation equipment. The method is characterized by firstly selecting the algorithm for energy efficiency calculation according to the characteristics of the fluid transportation equipment and establishing an input parameter list and an output parameter list; then carrying out energyefficiency calculation modeling by utilizing the parameters in the input parameter list according to the selected algorithm for energy efficiency calculation, thus establishing energy efficiency calculation models, and returning the results of energy efficiency calculation to the output parameter list; and then adopting one or more functions to ensure the established energy efficiency calculationmodels to correspond to the fluid transportation equipment. The method has the following technical effects: the energy efficiency calculation models are utilized for calculation, thus improving the efficiency of energy efficiency calculation; and important means are provided for energy efficiency management of the fluid transportation equipment.

The invention relates to a plant growth self-feedback learning cultivation method based on artificial intelligence, and the method comprises the steps: continuously collecting two-dimensional and three-dimensional images of each stage of plant growth, and continuously collecting plant growth environment data; inputting the collected two-dimensional and three-dimensional images into a plant growth model, optimizing the plant growth model, and obtaining current plant growth condition data; meanwhile, according to the obtained current plant growth condition data, based on the set plant growth parameters, configuring the growth environment of the plants in all the plant cultivation boxes in all the growth stages; and training a classification algorithm based on the obtained plant growth condition data and plant growth environment data to obtain a production parameter tuning model and tuning growth parameters. And obtaining the optimal plant growth parameters after multiple iterations. According to the method, real-time statistics and analysis can be carried out on the plant from the initial seedling planting stage to the flowering result, and plant growth research is automatically carried out in a self-feedback mode.

The invention provides an SLAM system based on data association and error detection and a method for constructing data association, wherein the system comprises a frontend module and a key frame database. The process for constructing data association by the frontend module comprises the steps of S100, receiving a common frame and a predicated pose Tpredict of the common frame; S200, searching n candidate frames which are similar with the common frame from a key frame database, and classifying the n candidate frames to a global candidate frame set and a local candidate frame set; S300, calculating a pose transform relation Tpc between each candidate frame and the common frame, and using Tpc as an association result; S400, performing error testing on the association result according to classification of the candidate frame for filtering error association results; S500, determining whether a correct association result is generated after S400, and if yes, returning for executing the S100,and otherwise, executing S600; and S600, creating a new key frame according to the historical common frame in a temporary region, the predicated pose, the common frame and the predicated pose thereof.

The invention discloses a method and device for obtaining a posture classification model. The method comprises the following steps carrying out the posture recognition of an image accommodating a preset posture through a key point detection model, and obtaining a posture label of the image accommodating the preset posture; and performing model training according to the image accommodating the predetermined attitude and the attitude label to obtain an attitude classification model. The attitude classification model obtained by using the method is simpler in network structure and lower in requirement on calculation resources, and can be operated on mobile equipment in real time; the human body posture recognition process of the posture classification model is only a posture recognition matching process, human body skeleton key point detection is not needed, the recognition process is simple, the recognition efficiency is high, and the processing time of a single-frame image is shortened.