Patents
Literature

871results about How to "Improve robustness" patented technology

Convolutional neural network based crowd density distribution estimation method

The invention relates to a convolutional neural network based crowd density distribution heat map generation method, which comprises the steps of dividing a crowd picture set into a training sample set and a test sample set, performing crowd label image segmentation by using convolutional neural network architecture, and carrying out number regression by using a convolutional neural network; correcting a density map through a multi-scale template operation, generating a crowd density distribution heat map according to the corrected density map and the regression number, and completing crowd density distribution estimation. According to the invention, deep characteristics of an image are extracted by using a powerful learning ability of the full convolutional neural network so as to perform accurate crowd segmentation, and low efficiency and blindness of density calculation of a traditional method for full image characteristics are overcome; a crowd near-far perspective effect is overcome to a certain extent through multi-scale template correction; and mapping is performed in allusion to the estimated number of people, lateral comparison can be performed on heat maps of different cameras, the method is applicable to various crowd scenes, and the crowd density distribution heat map can be acquired in real time.
Owner:ZHENGZHOU JINHUI COMP SYST ENG

Method for identifying handwritten numbers based on convolutional neural network and support vector machine

The invention discloses a method for identifying handwritten numbers based on a convolutional neural network (CNN) and a support vector machine (SVM). The combination of a convolutional neural network and a support vector machine increases the accuracy of identifying handwritten numbers. The method includes the following steps: enlarging a handwritten number image training set; conducting normalization; establishing two CNNS; training the two CNNs; establishing the SVM; remaining the alternating part of convolutional layers and pooling layers at the front edge of fully connected layers of the two CNNs, putting the fully connected layers of the two CNNs in serial connection and connecting the fully connected layers to the SVM to obtain a combined network; training the combined network; testing the handwritten number image testing set to obtain the result of the identification of the handwritten number. The method yields accuracy as high as 99.60%. According to the invention, the method obviates the need for complex pre-processing and is more adaptive and stable. The method also has high identification accuracy and is more reliable and robust. The method is applied to handwritten number identification in the fields of finance, postal service, data statistics, etc.
Owner:XIDIAN UNIV

Cascade regression-based face key point positioning method

The invention relates to a cascade regression-based face key point positioning method. The cascade regression-based face key point positioning method includes the following steps that: 1) a large number of face image data are acquired, and an initial key point position is marked; 2) the face image data are trained, and a coarse regression machine can be obtained through learning, and then with the output of the coarse regression machine adopted as input, a fine regression machine can be obtained through learning; and 3) face image data to be recognized are given, and the initial shape of a face is regressed to the vicinity of a real shape through the coarse regression machine, and then, with the output of the coarse regression machine adopted as input, the precise coordinates of a face key point can be obtained through the fine regression machine. According to the cascade regression-based face key points positioning method of the invention, a coarse-to-fine cascade regression algorithm is adopted, and a large number of samples are learned, and multi-feature fusion and multiple-regression machine fusion are realized, and therefore, the speed and the robustness of the algorithm are improved greatly, and the face key point can be excellently positioned under the situations of occlusion, low lightness and poses such as side faces, and the accuracy and the speed of face key point positioning can be effectively improved.
Owner:BEIJING KUANGSHI TECH

Target counting method and system based on double-attention multi-scale cascade network

ActiveCN110188685AImprove target count biasImprove robustnessCharacter and pattern recognitionNeural architecturesRobustificationPyramid
The invention discloses a target counting method and system based on a double-attention multi-scale cascade network. The target counting method comprises the following steps: inputting an image subjected to graying processing into an initial module of the double-attention multi-scale cascade network for initial feature extraction to obtain an initial feature map; inputting the initial feature mapinto a first branch network of the cascade network to obtain a low-level detail feature map and a high-level semantic feature map; performing channel attention transformation on the feature map to generate global feature information; inputting the initial feature map into a spatial attention pyramid structure of a second branch network in the cascade network to generate multi-scale features of theimage; fusing the multi-scale features and the global feature information to obtain a fusion result; performing feature extraction on the fusion result to generate an estimated target distribution density map; and carrying out pixel summation on the target distribution density map to obtain an estimated target counting result. The target counting method can effectively realize accurate target counting in a complex scene, and has good robustness and generalization.
Owner:YANSHAN UNIV

Automatic parking method based on fusion of vision and ultrasonic perception

ActiveCN110775052AImprove robustnessImprove obstacle avoidance reliabilityControl devicesObstacle avoidanceParking space
The invention relates to an automatic parking method based on fusion of vision and ultrasonic perception. The method comprises the following steps that visual parking space detection is carried out, ultrasonic radar parking space detection is carried out, two parking space detection results are fused, and parking space tracking and parking space display are carried out; a target parking space is determined, and a parking speed and a parking track are planned in combination with a current scene map; according to the visual detection and the ultrasonic radar detection, obstacles at a driving area around a vehicle and an edge of the driving area are detected, and the obstacles are subjected to motion estimation; the probability of collision of the vehicle itself and the obstacles is calculated, and if the collision probability is greater than a preset threshold value, an obstacle avoidance function is triggered; and parking track tracking is carried out through a low-speed controller, gear control, vehicle speed control, steering wheel steering control and obstacle avoidance control along the parking track are achieved, and automatic parking is completed. According to the method, thedetection effect of the parking space and the obstacles around the vehicle is improved, the robustness is improved, a motion obstacle track estimation module is added, and the obstacle avoidance reliability is improved.
Owner:ZHEJIANG LEAPMOTOR TECH CO LTD

Fatigue driving detection method and system based on convolutional neural network

The invention discloses a fatigue driving detection method and a fatigue driving detection method system based on a convolutional neural network and belongs to the technical field of image processingand mode recognition. The method comprises the steps of firstly, collecting a two-dimensional face image of a driver in a driving state, classifying stepwisely according to the fatigue degree, and establishing a fatigue driving image library; secondly, constructing the convolutional neural network containing a data layer, a convolutional layer, a pooling layer, a connection layer and a classification layer; thirdly, iteratively training the constructed network with image data and labels in the fatigue driving image library as input of the convolutional neural network by use of a back propagation algorithm so that loss function values output by the network decrease gradually and are restrained; and fourthly, inputting a test sample of the face image of the driver in the driving state, identifying the test sample by use of a trained convolutional neural network model so as to implement detection classification of the fatigue degree of the face image of the driver. According to the fatigue driving detection method and the fatigue driving detection method system based on the convolutional neural network, compared with a conventional machine learning method, identification and classification effects are obviously improved, and a feasible concept is provided for real-time monitoring of fatigue driving.
Owner:NANJING UNIV OF POSTS & TELECOMM

Graph convolutional neural network model and vehicle trajectory prediction method using same

The invention discloses a graph convolutional neural network model and a vehicle trajectory prediction method using the same. The model is composed of an encoder module, a spatial information extraction layer module and a decoder module. The method comprises the following steps: firstly, sampling a predicted vehicle and surrounding vehicles in a traffic scene at a frequency of 5Hz, and collectingposition coordinates and kinetic parameters of each vehicle sampling point, including horizontal and longitudinal coordinates, horizontal and longitudinal vehicle speeds and accelerations; calculatingcollision time TTC between the predicted vehicle and surrounding vehicles according to the coordinates and speeds of the predicted vehicle and the surrounding vehicles, and judging vehicle behaviors;inputting each historical track of the vehicle containing the information into the model, encoding time sequence interaction features in the track, extracting spatial features, summarizing the features into context vectors, and inputting the context vectors into an LSTM decoder to generate future track coordinates of the vehicle. According to the method, the problem that feature information generated by vehicle interaction cannot be obtained by using a traditional recurrent neural network is solved, and the prediction precision of the vehicle trajectory is greatly improved.
Owner:JIANGSU UNIV

Integrated power quality adjustment control method and device

The invention relates to a unified power quality control approach and the device thereof. The device comprises a parallel converting-bridge, a series converting-bridge, a DC power source as well as a third converting-bridge. The approach comprises the steps that: the compensation for a variable value to gain a power grid; the compensation for a voltage distortion as well as a target current value and the change ratio of both the voltage distortion and the target current value are input taken as a controller; the fuzzy control rule base and membership function parameter to be adjusted are selected as antibodies; the optimized important parameter of the fuzzy controller is realized through an immunity genetic algorithm; a SVPWM drive signal is generated and drives the series/parallel active filter power component; the difference value of the required capacitor voltage value and the actual capacitor voltage tested by a Hall sensor is applied to a PI algorithm; the SVPWM drive signal is generated again to drive the third converting-bridge power component. The invention combines the genetic algorithm, the immunity algorithm and the fuzzy algorithm; therefore, the robustness and the dynamic-static performance of the controller have been remarkably enhanced.
Owner:TIANJIN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products