35results about How to "Overcome the disadvantage of large amount of calculation" patented technology

The method for quickly detecting vehicles in aerial traffic video, the steps are as follows: step 100, adopt global motion estimation in the space-based encoding part, determine the global motion vector of the background; step 200, calculate the residual value according to the global motion vector, and segment the background area and the motion area ; Step 300, judge whether they are all background areas, transfer to the next frame for the images that are all background areas, and perform step 100; otherwise, for the motion area of ​​the image, perform step 400; step 400, first determine an offset on the ground part Low but still able to correctly segment the adaptive gradient threshold of each object, and perform preliminary marker extraction; then introduce the two parameters of area and catchment depth, and further screen the extracted markers to determine the final marker points; then use the marker points as The minimum value of the area is divided by VS watershed; finally, the area is merged according to the texture information of the area; step 500, the shadow is detected in the HSV color space, the false target is filtered out, and the vehicle is finally detected. The invention solves the problem that the calculation amount is large when decompressing the aerial image and detecting the moving target, and it is difficult to meet the requirements of real-time and robustness at the same time.

The invention belongs to the technical field of traffic surveillance and video processing and particularly discloses a video-based method for rapidly detecting transit vehicles. Based on a background difference method, a gradient frame difference method and a frame difference method, the method can simultaneously eliminate the daytime shadows, respiration effect, night headlamp ground-reflected light and other impacts, accurately detect the position of a moving object and judge the lane, where the vehicle is located. According to test results, the method of the invention can overcome the defect that the calculation load is higher in the conventional method and meet the requirement for the real-time performance in the video vehicle detection carried out by an electronic police; and the method is capable of accurately detecting the vehicles in the sight, and has better robustness against the interference, such as light changes, shadows and the like.

The invention discloses a linear frequency modulation signal parameter valuation method based on power spectrum. The method comprises that value-to-be-estimated linear frequency modulation signal power spectrum P (x) is detected; a shape feature CO (P (x)) of the linear frequency modulation signal power spectrum P (x) is extracted; a slope beta of a rising edge of the shape feature CO (P (x)) is calculated, and a frequency modulation slope k of the linear frequency modulation signal is calculated according to the relation between the frequency modulation slope k and the slope beta of the rising edge; a frequency difference delta f is calculated according to the relation between the frequency difference delta f of a peak frequency f1 and a starting frequency fs of the rising edge of the shape feature and the frequency modulation slope k; the starting frequency fs and an end frequency fe of the linear frequency modulation signal are calculated, and bandwidth B is calculated; and pulse width Tau is calculated according to the relation between the pulse width Tauand the bandwidth B. The defects in the prior art that influence of noise on linear frequency modulation signal value estimating result in a low signal noise ratio environment is large and calculation amount is large are overcome, and therefore value estimation of the linear frequency modulation signal parameter is achieved in unknown prior information conditions.

The invention relates to the field of sonar image processing, in particular to an image registration method of acoustic and visual three-dimensional imaging with an underwater vehicle, which overcomes target sonar image rotation due to swing of the underwater vehicle, and eliminates adverse effects of three-dimensional reconstruction distortion. The method comprises the following steps of (1), detecting a rotation angle of a target image, (2), correcting rotation of the target image, and (3), reconstructing a corrected image sequence. The method can correct sonar image sequences in a self-adapting manner without any prior knowledge to allow the sonar image sequences to be aligned axially, overcomes the defect that a reference image is required in the traditional registration method, is a quick automatic registration method, and is small in computation amount.

The invention relates to a low noise GPS (Global Positioning System) carrier wave tracking method based on an RLS (Recursive least square filter). The low noise GPS carrier wave tracking method comprises the following steps of: measuring an instantaneous phase difference of a local carrier wave and a tracking satellite by a carrier wave tracking loop by utilizing a table look-up type second quadrant arc tangent phase discriminator; introducing the RLS to filter measurement noises; meanwhile estimating a frequency difference according to phase difference variation; utilizing the frequency difference and the phase difference as frequency discrimination input and phase discrimination input of a loop filter with a second order lock frequency loop assistant in a three-order phase locking loop; filtering by the loop filter to output a carrier wave control adjustment variable with high precision and low noise, and correspondingly adjusting carrier wave NCO (Numerical Control Oscillation) to control the local carrier wave so that a receiver keeps phases and frequencies of the local carrier and a tracking satellite carrier wave consistent. In the invention, the low noise GPS carrier wave tracking method based on the RSL ensures that the GPS carrier wave tracking loop can work stably under the condition of low signal to noise ratio and certain dynamic state, keeps the phase not to overturn or skip cycles and ensures the carrier wave tracking precision.

The invention discloses a texture image segmentation method combined with sparse neighbor propagation and rapid spectral clustering, and the method mainly solves the problems that in an existing texture image segmentation method, the segmentation accuracy is low and the computation complexity is high. The method comprises the steps that 1, an image to be segmented is input; 2, parameters are set; 3, a grayscale-probability square matrix is generated; 4, the number of data points is counted; 5, a sparse similarity matrix is built; 6, data representative points are selected; 7, clustering is performed on the data points; 8, the image to be segmented is marked; 9, a segmented image is output. Compared with some existing texture image segmentation technologies, the texture image segmentation method combined with sparse neighbor propagation and rapid spectral clustering can better maintain the region consistency of the texture image, the texture image with high segmentation accuracy and good segmentation stability can be obtained, and the computation complexity is low.

The invention discloses an information source number estimation method for a polarization sensitive array. The method includes the following steps that spatial electromagnetic wave signals are received through a polarization sensitive array formed by N polarization sensitive array elements, and receiving data are obtained; a covariance matrix is calculated according to the data received by the polarization sensitive array; feature value decomposition is conducted on the covariance matrix, 2 N feature values and feature vectors corresponding to the 2 N feature values are solved, and the 2 N feature values are sorted in a descending mode from big to small; feature removal treatment is conducted on the covariance matrix through the sorted front N feature values, corresponding N new covariance matrixes are obtained, and a projection matrix is built through sub spaces formed by the feature vectors corresponding to the sorted rear N feature values; projection on the projection matrix of the N new covariance matrixes is solved respectively; the information source number is estimated according to a projection structure criterion function. According to the information source number estimation method for the polarization sensitive array, the information source number is estimated through a noise sub spatial dimension feature of the polarization sensitive array, the calculated amount is reduced, and the cost is saved.

The invention discloses an intelligent real-time material position coordinate recognition and random point sampling method. The steps comprises: (1) establishing and correlation method of space coordinates constructed by a sensor and space absolute coordinates in an objective reality environment; (2) car body real-time data collection. (3) car body and cargo outline recognition; (4) a random point generation method in a cargo real-time coordinate; and (5) a human-computer interface.

The invention discloses a teaching video labeling method based on collaborative filtering, which mainly solves the disadvantage of low accuracy rate of teaching video labeling in the prior art. The implementation steps are as follows: input the teaching video, and extract the subtitle key frames of the teaching video according to the subtitles to obtain D key frames; use the optical character software to extract the subtitles of the D key frames, and modify and delete the obtained subtitles , get D text documents; use D text documents combined with Gibbs sampler to segment the teaching video, and divide the teaching video into M shots; in the M shots, mark some shots, and then use the collaborative filtering method to calculate the marked shots The cosine similarity between the unmarked shot and the first 5 words with high cosine similarity is selected to mark the unmarked shot. Since the present invention considers the subtitle information in the teaching video, it can describe the teaching video more effectively, improves the marking accuracy of the teaching video, and can be used in video teaching.

The invention discloses an image spectral clustering method based on quick selection of landmark points and mainly aims to solve the problems that existing image spectral clustering methods are low inclustering precision and high in calculation complexity. The method comprises the steps that (1) all images to be subjected to spectral clustering are read; (2) neighbor images of the images to be subjected to spectral clustering are calculated; (3) the landmark points are selected; (4) a feature sparse representation matrix of the images to be subjected to spectral clustering is calculated; (5)a relevant matrix of the images to be subjected to spectral clustering is calculated; (6) a right singular feature matrix of the sparse representation matrix is calculated; and (7) recognition clustering is performed. Compared with some existing image spectral clustering technologies, through the method, a sparse representation error of the images can be lowered, the accuracy of a spectral clustering result is improved, and the calculation complexity is low.

The invention discloses a blockchain asset transaction transfer method based on a weighted threshold signature algorithm. The specific implementation steps include: 1. Generating a public key-private key pair of the group owner; 2. Distributing weighted private key components; 3. Generating a message to be signed ;4. Several group members form a signature group; 5. The group owner judges whether the signature group meets the threshold condition; 6. Calculates the signature private key component; 7. Calculates the temporary public key; 8. Sends the signature component; 9. Synthesizes complete Signature; 10. Verify signature; 11. Exit the system. The invention adopts a threshold signature algorithm, reduces the risk of private key leakage, and ensures the security of blockchain asset transaction transfer. Using the weighted secret sharing technique based on the Chinese Remainder Theorem (CRT), the efficiency of the signature generation algorithm is improved.