10 results about "Optical flow" patented technology

The invention discloses a device and method for monitoring and early warning of flood damages based on multi-resource integration. The device comprises a hydrological actual state module, a flow forecast module, a weather forecast module, a radar analysis module, a forecast integration module and a module for product display and early warning release. The method comprises the steps that (1) the hydrological actual state module reads and monitors monitoring point data; (2) the weather forecast module reads input data and parameters to carry out weather mode forecasting; (3) the flow forecast module carries out flow forecasting according to the weather mode forecast and the monitoring point data provided by the hydrological actual state module; (4) the radar analysis module reaches basic data of a radar and estimates precipitation and optical-flow extrapolation; (5) the forecast integration module integrates and outputs information of the flow forecast module and the radar analysis module; and (6) the module for the production display and the early warning release completes the product display and the early warning release. The device and the method consider accuracy of radar extrapolation forecasting in forecasting of the position of a precipitation system and also consider capacity of numerical mode forecasting, so that limitations of the single method are avoided.

The invention provides a behavior recognition method and system based on an attention mechanism double-flow network, and belongs to the technical field of behavior recognition, and the method comprises the steps: dividing an obtained whole video segment into a plurality of video segments with the same length, extracting an RGB image and an optical flow gray-scale image of each frame of each videosegment, and carrying out the preprocessing of the RGB images and the optical flow gray-scale images; carrying out random sampling on the preprocessed image to obtain an RGB image and an optical flowgrayscale image of each video clip; extracting appearance features and time dynamic features of the sampled images by using a double-flow network model introducing an attention mechanism, fusing the appearance features and the time dynamic features according to the types of a time domain network and a space domain network respectively, and performing weighted fusion on a fusion result of the timedomain network and a fusion result of the space domain network to obtain an identification result of the whole video. According to the invention, the video data can be fully utilized, the local key features of the video frame image can be better extracted, the foreground area where the action occurs is highlighted, the influence of irrelevant information in the background environment is inhibited,and the behavior recognition accuracy is improved.

The invention discloses a real-time gesture recognition method which comprises the steps of (1) decomposing an obtained gesture video into image sequences sorted in a chronological order and preprocessing obtained images and then carrying out hand-region segmentation, (2) extracting a hand shape feature of a hand region in each image and using an SVM support vector machine to identify the hand shape feature as a corresponding gesture value, (3) combining the gesture value of each image and a direction feature of a motion trajectory obtained by an iterating LK pyramid optical flow algorithm asa feature vector of each dynamic gesture image, (4) carrying out loop execution of steps (2) and (3) with a loop end condition that all images of a current video are processed so as to obtain a complete set of feature vector sequences, (5) creating a gesture template library, (6) carrying out optimization DTW match on the obtained feature vector sequences and all templates in a template library, calculating the degree of distortion of the match, wherein recognition is failed if the degree of distortion is larger than a distortion threshold, and a recognition result is outputted if the degree of distortion is smaller than the distortion threshold.

A method of unlocking an electronic device based on motion recognitions generates motion information by detecting a user's motion, calculates optical flows based on the motion information, converts the optical flows into a code by sorting and grouping the optical flows based on predetermined reference directions, generates a comparison result indicating whether the code is consistent with a predetermined password by comparing the code with the predetermined password, and determines whether to unlock the electronic device based on the comparison result.

The invention provides an image processing device and an image processing method. The image processing device comprises at least one image capturing device and a processor. The processor receives the first image and the second image from the image capturing device and identifies an overlapping region between the first image and the second image. The processor analyzes the overlapping region to obtain a plurality of feature points, a first optical flow adjustment region and a second optical flow adjustment region, dividing the first optical flow adjustment area and the second optical flow adjustment area into a plurality of triangular blocks by using the feature points as vertexes, the corresponding triangular blocks in the first and second optical flow adjusted regions are mixed to form a stitched region, and the stitched region and a plurality of non-overlapping regions in the first and second images are combined to generate a stitched image. Therefore, the operation complexity is greatly reduced, the better image quality can be achieved, and the method is suitable for various real-time applications such as real-time operation of image splicing, three-dimensional image and image generation.

The invention discloses a zinc flotation working condition identification method based on long-time-history depth features, and the method comprises the following steps: firstly, employing a 3D convolutional network as a basic network, and simulating an optical flow network through a knowledge distillation method by employing a part of structure of an RGB flow network, so that the RGB flow network can learn the motion information of the optical flow without extracting the optical flow during testing; segmenting a single video, performing frame-level feature extraction on each segment by using a trained RGB flow network, and inputting the extracted frame-level features of each segment into an LSTM network to further extract video-level global spatial-temporal features; and supplementing a 2D convolutional network for the network to extract supplemented appearance features, splicing the global spatial-temporal features and the enhanced appearance features together, and inputting the spliced features into a multi-layer perceptron to carry out final working condition identification. According to the invention, the advantages of the convolutional neural network and the recurrent neural network are combined, and the zinc flotation working condition can be quickly and accurately identified, so that dosing is effectively guided.

The invention relates to the technical field of computer vision tracking, in particular to a fast human body searching method based on motion segmentation, comprising the following steps: step S1, motion segmentation is carried out by using an optical flow algorithm, a human body candidate region frame is determined, and the candidate region frame is extended for a distance upward; step S2, a facedetection algorithm of a small area is carried out in the candidate area frame, and it is judged whether a face exists or not. At the same time, the invention also discloses a fast human body searching system based on motion segmentation. The invention has the beneficial effects of greatly reducing the area executed by the classifier and improving the classification accuracy of the classifier.