669 results about "Correlation filter" patented technology

The invention discloses a scale adaptive target-tracking method based on a depth characteristic kernel correlation filter. The method comprises the following steps: inputting an image into a pre-trained convolution neural network, and extracting depth convolution features; tracking a target, and estimating the position and scale of the target through the trained model; training the kernel correlation filter according to the currently detected target position and scale; and updating the kernel correlation filter through employing an adaptive high-confidence model updating method. According to the invention, the depth convolution features are extracted, and the adaptive scale estimation method and the adaptive high-confidence model updating strategy are improved, thereby improving the target tracking robustness under the conditions of complex scenes and appearance changes. The method can achieve the high-efficiency and accurate processing of the scale change of the target. In addition, the adaptive high-confidence model updating strategy is employed, so the model tracking drift is reduced as much as possible.

The invention provides an MS-KCF (MobileNet-SSD-kernelized correlation filters) based rapid and stable detection method of a human face in an image sequence. To solve the problem about detection of the human face with larger angle change and very severe shielding in the image sequence, a new automatic DTD (detection-tracking-detection) mode integrating rapid and accurate target detection model MSand rapid tracking model KCF is provided, that is, an MS-KCF human face detection model is provided. The method comprises steps as follows: S1, establishing an MS detection network; S2, detecting a target by use of the MS network; S3, updating the tracking model to predict the position of the next frame of human face target; S4, updating an MS detection network after tracking several frames, and performing redetection and positioning on the human face target; S5, comparing and analyzing experiment results. Experiments prove that the MS-KFC model guarantees the detection stability of the humanface with larger angle change and very severe shielding in the image sequence, and besides, the detection speed is greatly increased.

A methodology is described to reduce the complexity of filters for face recognition by reducing the memory requirement to, for example, 2 bits/pixel in the frequency domain. Reduced-complexity correlations are achieved by having quantized MACE, UMACE, OTSDF, UOTSDF, MACH, and other filters, in conjunction with a quantized Fourier transform of the input image. This reduces complexity in comparison to the advanced correlation filters using full-phase correlation. However, the verification performance of the reduced complexity filters is comparable to that of full-complexity filters. A special case of using 4-phases to represent both the filter and training/test images in the Fourier domain leads to further reductions in the computational formulations. This also enables the storage and synthesis of filters in limited-memory and limited-computational power platforms such as PDAs, cell phones, etc. An online training algorithm implemented on a face verification system is described for synthesizing correlation filters to handle pose/scale variations. A way to perform efficient face localization is also discussed. Because of the rules governing abstracts, this abstract should not be used to construe the claims.

The invention provides a kernel correlation filtering target tracking method based on feature fusion and Bayesian classification. The method is characterized by comprising steps: firstly, the position and the scale information of an initial frame target are given; then, a standard target appearance model, a standard correlation filter and a standard color Bayesian model are built or updated; based on the target center point of a former frame, a search area is extracted; and a correlation filter of a Gaussian kernel is used for realizing target displacement estimation, the color Bayesian model is used for realizing target scale estimation, and the tracking result of the current frame is obtained further. Through sequentially processing each frame of video image, tracking of a moving target in the video is realized. The target tracking method can effectively solve the problem of precisely positioning the moving target in the video and can realize target scale estimation. In multiple challenging environments, the target can be tracked effectively, certain robustness is realized, and the tracking algorithm precision is improved.

The invention relates to a robust target tracking method based on deep learning and multi-scale correlation filtering. The tracking process is divided into a target location part and a scale selection part. In the target location part, the position of a target is located through a convolutional neural network and correlation filtering. In the scale selection part, a scale pyramid is used, and different scales are selected in a matching manner for targets through scale filtering. The multilayer characteristic of the convolutional neural network is taken as a representation model of targets, so the structural and semantic information of targets can be described robustly. Through use of the characteristics of correlation filtering, there is no need to train a classifier online, and the running speed of the algorithm is increased greatly. The idea of scale pyramid is adopted in scale, and correlation filtering matching is performed on targets of different scales to select the optimal scale. The method is of strong robustness to deformation, shading and scale change of targets.

The invention provides a relevant filtering opposite-thrust target tracking method with an adaptive scale. The method comprises: an initial position and an initial scale of a to-be-tracked target in a video frame are determined, and convolution feature graphs of different layers are extracted respectively by using the initial position as the center and using a deep convolutional neural network; for the extracted convolution feature graph of each layer, tracking is carried out by using a kernel-correlation filtering tracking method to obtain a tracking result; all tracking results are combined by using an adaptive hedging algorithm to obtain a final tracking result as a final position of the to-be-tracked target, so that the to-be-tracked target in the video frame can be localized; after obtaining of the final position of the to-be-tracked target, a final scale of the to-be-tracked target is estimated by using a scale pyramid strategy; and after obtaining of the final position and the final scale of the to-be-tracked target, a to-be-tracked target image block is extracted based on the final scale by using the final position as a center and each kernel-correlation filtering tracking method is trained again to update a coefficient and a template.

Ambience signal components are obtained from source audio signals, matrix-decoded signal components are obtained from the source audio signals, and the ambience signal components are controllably combined with the matrix-decoded signal components. Obtaining ambience signal components may include applying at least one decorrelation filter sequence. The same decorrelation filter sequence may be applied to each of the input audio signals or, alternatively, a different decorrelation filter sequence may be applied to each of the input audio signals.

Disclosed herein are a method for tracking object in video in real time in consideration of both color and shape and an apparatus for the same. The apparatus may generate at least one of a correlation filter model and a color probability density model based on an input image, determine an initial position and an initial size of a target to be tracked based on a correlation filter response value calculated by applying the correlation filter model to the input image, calculate a joint probability based on the correlation filter response value and a color object probability that is calculated based on the color probability density model, and update an object model corresponding to the target to be tracked based on a final position and a final size of the target to be tracked, which are determined by applying the joint probability to the initial position and the initial size.

The invention discloses a KCF (Kernel Correlation Filtering) target tracking method employing CNN (Convolution Neural Network) in integrated manner and belongs to the technical field of image processing. First, a first frame of a video is read, characteristics of a target is extracted and idealization output is given at the same time; a KCF template is obtained through training. Then, a next frameof image is read, a tracking result of a KCF algorithm is calculated and a KCF response map and a target result KCF_Box are obtained; and a PSR value of the KCF algorithm is calculated. Whether the PSR value is greater than an algorithm threshold value or not is judged. If the PSR value is greater than an algorithm threshold value, calculation employing a GOTURN algorithm is not needed and KCF_Box as the result of the KCF algorithm is taken as the tracking result of the current frame. If the PSR value is not greater than an algorithm threshold value, calculation employing the GOTURN algorithmis performed and a tracking result GOTURN_Box of the GOTURN algorithm is taken as the tracking result of the current frame. Finally, template update of the KCF algorithm and network input update of the GOTURN algorithm are performed. According to the invention, a side lobe ratio is taken as a bridge and an integration method for the KCF algorithm and the GOTURN algorithm is proposed, so that theaccuracy of the target tracking result is ensured.

A receiver employs a pre-correlation filter to determine the precise timing of, for example, a PRN code on the quadrature channel of a received signal, using an image of the average chip shape that the filter forms for a PRN code on the in-phase channel. The image is expressed as a time series of complex power measurements along the length of a single chip. The averaging process retains the detail of the composite in-phase signal (direct plus multipath signals) while reducing the level of signal noise by an amount proportional to the length of the averaging process. An analysis of the image reveals that there is, in the in-phase channel signal that is averaged, information from the quadrature channel signal. The quadrature channel signal information produces, in the image of the average chip shape of the in-phase channel PRN code, a “wiggle” that corresponds to the timing of the chips of the quadrature channel PRN code. The receiver detects the chip edges of the quadrature PRN code directly from an analysis of the high frequency phase modulations of the complex vector of samples that represents the average chip shape. Using GPS signals, the receiver detects the P-code transitions by synchronizing to the 10.23 MHz phase modulations in the complex vector of samples that represent the averaged chip shape. The receiver uses the detected P-code transitions and, more particularly, the P-code transitions that are closest to the C/A code transitions, to produce P-code phase information that the receiver uses pseudorange calculations to remove biases associated with timing differences between the transmission of the in-phase and quadrature PRN codes.

The related echo eliminator comprises: a linear predicted decoder to decode data into remote speech signal and extract the short activating signal and linear prediction coding parameter, a decorrelation filter to receive the predicted coding parameter and error signal for decorrelation and generate decorrelated error signal, an adaptive filter to receive all of the remote speech and short activating signal and decorrelated error signal and generate the estimated echo, and a subtracter to generate the error signal. This invention combines with LPC device, reduces algorithm complexity, and prompts algorithm convergence.

The invention provides a multi-target tracking method based on time series multi-feature fusion. The method of the invention includes that: the category and the candidate frame of the tracking targetare obtained according to the multi-target detection algorithm; a convolution network and a correlation filter are used to compute a moving prediction center point and screen candidate frames; appearance similarity score is calculated; the motion similarity score is calculated; an interactive feature similarity score is calculated; the selected candidate frame is converted into a tracking frame ofthe current frame image, and the characteristic information of the tracking object is updated ; a moving prediction center point of a tracking target that is not matched to a candidate frame is calculated and the candidate frame is screened; unmatched candidate boxes are associated for existing tracking targets to construct new tracking targets; the overlap degree of each tracking object is calculated by using the intersection-union ratio; the lost targets in the multi-frame images are identified as the lost targets. Compared with the prior art, the invention improves the tracking accuracy.

The invention relates to a multi-characteristic layered fusion correlation filtering robustness tracking method. In order to improve the robustness of target tracking, aiming at the multi-characteristic fusion problem in correlation filtering tracking, a multi-characteristic layered fusion strategy is provided. HOG characteristics, CN characteristics and color histogram characteristics are extracted from a target area and a surrounding background area respectively. A self-adaption weighting fusion strategy is adopted for integration of characteristic response diagrams of the HOG characteristics and the CN characteristics. A fusion result of a layer and the obtained characteristic response diagrams based on the color histogram characteristics are subjected to second-layer integration, a fixed parameter fusion strategy is adopted for fusion of the characteristic response diagrams. According to the method, under the premise that the tracking accuracy is guaranteed, the tracking robustnessis superior to other algorithms. When multiple characteristics are adopted for a correlation filtering tracking algorithm, the layered fusion strategy has certain reference.

The invention relates to an echo canceller, an echo cancellation method and the double-end talking detection system thereof, wherein, the echo canceller decodes a data stream from the network end into a far-end voice signal and extracts a short-time excitation signal as well as a linear prediction coding coefficient by utilizing a linear prediction decoder. Based on the linear prediction coding coefficient, the echo canceller carries out a decorrelation process to a residual signal by a decorrelation filter, and executes a normalized cross correlation process to the short-time excitation signal and the decorrelation residual signal by utilizing the double-end talking detector to generate the normalized cross correlation detection variable, hence, the echo canceller detects the existence of the double-end talking and then further controls the renewal of the self-adaptive filter. The invention is capable of detecting the existence of the double-end talking by the decorrelation process as well as the normalized cross correlation process to the far-end voice signal and the residual signal, thus obviously decreasing the calculation complexity of the double-end talking detection and the storage space consumption.

The invention discloses a correlation filtering and deep twin network-based robust long-range target tracking method, and relates to a computer vision technology. According to the method, correlationfiltering and a deep twin network are combined under a uniform tracking framework, so that challenges of target occlusion and view disappear in long videos can be effectively processed. In the tracking method, a D-expert and C-expert-based expert assessment mechanism is capable of effectively assessing and screening target candidate positions jointly generated by the correlation filtering and thedeep twin network so as to obtain optimum target tracking results, and the results are used for updating correlation filtering trackers, so that the condition that the correlation filtering trackers are updated by wrong samples is effectively avoided. The target tracking method is relatively robust for various challenges in long videos, and is capable of stably tracking targets for a long time.

A high-speed automatic multi-object tracking method with kernelized correlation filters is provided. The method includes obtaining an image frame from a plurality of image frames in a video, extracting a foreground object sequence from the obtained image frame, and determining similarity between each foreground object of the extracted foreground object sequence and a tracked object. The method also includes calculating HOG features of the foreground objects with a lower similarity, obtaining training samples for each of the foreground objects with the lower similarity using a circular matrix, obtaining a classifier via a kernel method accelerated by FFTW, and obtaining tracking points using a sparse optical flow. Further, the method includes detecting object matching responses using a detection response function, performing multi-scale analysis for the object based on an optical flow method, and processing a next image frame of the plurality of image frames in the video until the video ends.

The invention provides a target tracking method based on multi-characteristic adaptive fusion and kernelized correlation filtering technology. The method comprises steps of according to target position and the dimension of the previous frame tracking, acquiring a candidate region of target motion; extracting histogram characteristics and color characteristics in the gradient direction of the candidate region, fusing the two kinds of characteristics, carrying out Fourier transform so as to obtain a characteristic spectrum and then calculating kernelized correlation; determining the position and the dimension of the target at the current frame, and acquiring a target region; extracting histogram characteristics and color characteristics in the gradient direction of the target region, fusing the two kinds of characteristics, carrying out Fourier transform so as to obtain a characteristic spectrum and then calculating kernelized self-correlation; designing the adaptive target correlation and training a position filter model and a dimension filter model; and using a linear interpolation method to update the characteristic spectrums and the related filters. According to the invention, the discrimination capability of the models is improved; robustness of the target tracking of the target in a complex scene and the appearance change is improved; calculation complexity is reduced; and tracking timeliness is improved.

The invention relates to a scale adaptive kernel correlation filter fast target tracking method. According to the scale adaptive kernel correlation filter fast target tracking method, separation of afront background is promoted and a scale estimation module is added through introducing a two-dimensional variable bandwidth Gaussian window into a KCF framework, scale variations of a target are estimated according to a matching result of feature points of an image block, and whether the target is sheltered is judged by combining displacement prediction based on the scale variations. Once the target is sheltered, the user considers that no scale conversion occurs at the moment and template update is not performed, further the tracking precision is enhanced in this way. The scale adaptive kernel correlation filter fast target tracking method has the advantages of simple and fast calculation, and can significantly improve the calculation efficiency on the basis of compatible scale adaptation.

The invention provides a moving target tracking method based on an image, and aims to the problem of tracking failure of the conventional tracking method based on a convolutional neural network duringsize or scale change of a moving target. According to the method, feature extraction is performed on the image through a pre-trained network by employing a deep learning technology, a target point istracked by employing a correlation filter method, the tracking accuracy is improved through acquisition of information of time correlation by employing a C3D network, and multi-scale tracking characteristics are provided through combination of a DenseNet network and a network structure of a feature pyramid network so that accurate tracking during change of the size or the scale of the moving target can be realized, and the method can be widely applied to tracking of the moving target based on the image.

The invention relates to a multi-layer convolution feature self-adaptive fusion moving target tracking method, and belongs to the field of computer vision. The method comprises the following steps: firstly, initializing a target area in a first frame of image, and utilizing a trained deep network framework VGG-19 to extract first and fifth layers of convolution features of the target image block,and obtaining two templates through learning and training of a related filter; Secondly, extracting features of a detection sample from the prediction position and the scale size of the next frame andthe previous frame of target, and carrying out convolution on the features of the detection sample and the two templates of the previous frame to obtain a response graph of the two-layer features; calculating the weight of the obtained response graph according to an APCE measurement method, and adaptively weighting and fusing the response graph to determine the final position of the target; And after the position is determined, estimating the target optimal scale by extracting the directional gradient histogram features of the multiple scales of the target. According to the method, the targetis positioned more accurately, and the tracking precision is improved.

The invention discloses a mechanism for positioning pointers of a watch and methods for positioning, resetting and calibrating the pointers and belongs to the technical field of the watch. The invention is technically characterized in that: an ordinary reflecting photoelectric sensor is arranged on the front surface of a movement of the watch and used for capturing pointer signals which are generated during the sweeping of the pointers; a microprocessor processes the pointer signals; a filtering module of the microprocessor performs correlation filtering on the pointer signals to improve the signal to noise ratio of the pointer signals; a pointer identification module performs statistic analysis and characteristic analysis on the pointer signals to determine the position of each pointer; and a motor control module controls the pointers to move according to the position of each pointer, so that the pointers can be automatically calibrated and reset. In the scheme, an ordinary process installation requirement is only required without designing of a special detection component, so that a technology for automatically calibrating the pointers of the watch can be popularized and applied on a large scale, the cost of products can be greatly reduced, and the invention has an important technical significance and a business value.

A radar receiver system includes a receiver, a processor including a Doppler Compensated Adaptive Pulse Compressor (DCAPC) algorithm, possible other intermediate processing and a target detector. The DCAPC algorithm processes samples of a radar return signal, applies Minimum Mean Square Error (MMSE), or alternatively matched filtering, to the radar return signal to obtain initial radar impulse response estimates, computes power estimates, estimates a range cell Doppler shift for each range cell, computes range-dependent filters, applies the MMSE filters, and then repeats the cycle for subsequent reiterative stages until a desired length−L range window is reached, thereby resolving the scatterer from noise and other scatterers.

A height dependent filter for a pointing device or digitizer tablet, a method of reducing jitter in position data of a pointing device, and a computer readable medium are provided. A method of stabilizing positional data in a digitizer tablet associated with a display device for reporting a position of a pointer includes detecting a current location of the pointer with respect to a coordinate grid defined on the tablet and outputting a position signal indicative of the current pointer location. Noise is filtered from the position signal based on at least one filter parameter. A distance between the pointer and a point on the tablet is determined, and an amount of jitter in the position signal is determined based on the determined distance. The at least one filter parameter is adjusted based on the estimated amount of jitter.

The invention discloses a real-time tracking method of a multi-channel kernelized correlation filter. The method includes the following steps: a training step of conducting ridge regression on a previous frame of object information to acquire a filtering template; a detection step of detecting the current frame of image with the acquired filtering template and outputting a filtering response; an updating step of real-time updating the filtering template and the appearance of an object. According to the invention, the method uses kernel function to fuse multi-channel characteristics, overcomes the selection limit of the multi-channel characteristics, and transforms the problem of linear optimization of ridge regression to the problem of non-linear optimization of a higher space through the kernel function, such that the filtering template with excellent robustness is constructed, the speed of a tracking machine is greatly increased, and tracking requirements of real world can be met.

The invention discloses an adaptive anti-occlusion infrared target tracking method based on multi-layer depth feature fusion. Firstly, a series of multi-layer depth feature maps of the same size and different levels are obtained; the multi-layer depth feature map is then converted from a time domain to a frequency domain according to correlation filtering, a filter is trained and the response mapis calculated according to the fast Fourier transformation, and then the multi-layer depth feature map is merged and dimensionally reduced according to the weighted fusion of the intra-layer features,so that the feature response map of different levels is constructed and the maximum correlation response value is obtained, which is the estimated position of the target; at last, that dense featuresof the target are extracted, and the response confidence of the target center position estimated by the depth convolution feature is obtained according to the maximum response value of the feature obtained by the correlation filtering; when the response confidence of the target center position is less than the re-detection threshold T0, the obtained target estimated position is evaluated by on-line target re-detection and the position of the target is adaptively updated according to the evaluation result.

The invention belongs to the field of infrared image processing, and relates to an infrared small unmanned aerial vehicle target detection and tracking method under a complex background. The method comprises the following steps of (S1) obtaining a training sample, and training a deep convolutional neural network as an unmanned aerial vehicle target detection network; (2) obtaining a to-be-detectedtarget image in real time, inputting the to-be-detected target image into the unmanned aerial vehicle target detection network in the step (1), and outputting an unmanned aerial vehicle target detection result; and (S3) tracking the unmanned aerial vehicle target output in the step (S2) by using a kernel correlation filtering rapid tracking method. According to the method, feature extraction is carried out by adopting the residual network based on batch regularization and random discarding, so that the training efficiency and the model robustness are improved. According to the method, the context features and the semantic features are fully combined, and multi-scale discrimination is carried out by using the multi-layer fusion feature map with fine granularity, so that the detection precision of a small target is effectively improved.