554 results about "Multiple frame" patented technology

The invention provides an improved multi-frame image target detection network, namely, TFA-Net (Temporal Fusion Attention Network), which is oriented to a welding seam defect detection task. According to the network, continuous multiple frames of images are used as input, and spatial features of each frame of image are extracted through a ResNet backbone network sharing parameters. On the basis, a multi-stage feature fusion module is fused and introduced, effective integration among features of different scales is realized, and the perception capability for small-size and weak-contrast defects is improved. In order to further capture dynamic information of the target in the time dimension, a time sequence modeling module is designed, modeling is carried out on a multi-frame feature sequence, and continuous features of the target changing along with time are extracted. And then, the network adopts a gating fusion mechanism to carry out adaptive weighted fusion on the static space features and the dynamic time sequence features, and the robustness and the discrimination capability of feature representation are enhanced. Finally, the fusion features are input into a decoder module, the category probability of the defects is predicted through a classification sub-network, anchor frame position offset is calculated through a frame regression sub-network, and accurate positioning and recognition of the multiple types of defects in the weld seam image are achieved. The network has the advantages of clear structure, strong generalization ability, high adaptability and the like, and is especially suitable for the problems of small target size, unclear texture, strong motion continuity and the like in a welding seam detection scene.

The invention discloses a multi-frame photoacoustic image reconstruction method based on optical flow alignment and depth feature fusion. The method comprises the following steps: S1, obtaining photoacoustic signal data; s2, reconstructing a photoacoustic cross-sectional image; s3, performing optical flow calculation and image alignment; and S4, training the deep feature fusion network. According to the method, through optical flow motion correction and a potential space learning mechanism, space-time information and complementary features in the aligned multiple frames of images are dynamically integrated, complementary information in the multiple frames of images is adaptively fused, noise is suppressed, and finally reconstruction of high signal-to-noise ratio and high spatial resolution images of biological tissues is achieved. Experimental results show that the method can significantly improve image quality, recover image distortion and detail loss caused by motion and noise, and provide a new effective scheme for promoting robust clinical application of a photoacoustic imaging technology.

The invention relates to a laser radar camera calibration method and device, and a medium. The method comprises the steps: collecting multi-frame laser radar point cloud data and synchronous corresponding image data in a construction scene; carrying out multi-frame point cloud fusion and dense reconstruction to generate a laser dense point cloud; visual sparse point cloud reconstruction is carried out, and cross-modal scale unification and space alignment are carried out on an initial visual point cloud obtained through reconstruction and the laser dense point cloud; generating a visual dense point cloud through three-dimensional Gaussian splashing; and performing registration on the laser dense point cloud and the visual dense point cloud by adopting a point-to-line iterative nearest point algorithm, and performing calculation to obtain an external parameter calibration matrix of the camera and the laser radar. Compared with the prior art, the method has the advantages of high precision, low cost, high stability and the like.

The invention discloses a recognition method, device and system for putting food materials in two hands and a storage medium, and belongs to the technical field of household appliances. The method comprises the following steps: acquiring multiple frames of first images; performing hand and food material detection and recognition on the multiple frames of first images to obtain hand and food material detection and recognition results corresponding to the first images in the multiple frames of first images; according to the hand and food material detection and recognition results corresponding to the first images in the multiple frames of first images, tracking the two-hand actions of the user, and determining whether the action of at least one hand in the two-hand actions of the user is food material putting or not; and under the condition that at least one of the actions of the two hands of the user is to put the food materials, determining a target shelf area in which the two hands put the food materials on the basis of a target image of a bounding box in which the hand and food material detection and recognition result in the multi-frame first image comprises the food material taking by the hand. By detecting the hands and the food materials and tracking the actions of the two hands of the user, accurate recognition of the food materials put into the two hands is achieved.

The invention relates to the technical field of medical image processing, in particular to an endoscopic surgery video real-time structure analysis method and system, and the method comprises the steps: carrying out the frame-by-frame semantic segmentation of real-time video data, and generating a segmentation mask of each frame of image; calculating a comprehensive quality score of the target frame based on the segmentation masks of the target frame and the previous frame of the target frame; analyzing a motion amount index between adjacent frames; generating the current length of a sliding time sequence window according to the comprehensive quality score and the exercise amount index, and fusing multiple frames of segmentation masks in the sliding time sequence window to generate a reference mask; extracting a key point from the reference mask, and obtaining a displacement vector of the key point from a previous frame of the target frame to the target frame; generating a pixel-level displacement field according to the displacement vector, deforming the segmentation mask of the previous frame of the target frame to the target frame, and generating a prediction mask of the target frame; and performing superposition display on the prediction mask and the image of the target frame. According to the scheme, the time sequence consistency and stability of the video semantic segmentation result can be enhanced.

The invention relates to a video generation method based on time sequence similarity, and is applied to the field of video generation. Specifically, the video generation method based on the time sequence similarity comprises the following steps: acquiring generated video information; generating a key video frame sequence with frame rate information based on the text information, wherein the key video sequence is separated by a plurality of frame marks; performing recursive interpolation processing on the key video frame sequence, fusing front and back key frame features through a preset attention mechanism, and generating an initial intermediate frame between adjacent key frames; extracting visual features of a plurality of key frames and intermediate frames included in the key video frame sequence, calculating cosine similarity of the visual features of adjacent frames, and adjusting intermediate frame generation parameters based on the similarity to optimize inter-frame coherence so as to obtain an optimized intermediate frame; and combining the key video frame sequence with the optimized middle frame to generate a complete video clip which is consistent with the text information and has a coherent inter-frame time sequence.

The invention discloses an ultrasonic image analysis method and an ultrasonic imaging system. The method comprises the following steps: acquiring multiple frames of ultrasonic images of the heart of a target object; automatically determining a myocardial contour of the multi-frame ultrasonic image; at least determining the image quality of the myocardial contour of the multi-frame ultrasonic image; determining a modification priority level of a myocardial contour region of each multi-frame ultrasonic image based on the image quality; and displaying at least one frame of ultrasonic image in the multiple frames of ultrasonic images and the corresponding identification information for identifying the modification priority. According to the ultrasonic image analysis method and the ultrasonic imaging system provided by the invention, the modification priority level of the ultrasonic image is determined, and the ultrasonic image and the corresponding identification information for identifying the modification priority level are displayed, so that the visual prompt for recommending the preferential modification frame is realized, and the operation and use efficiency of a doctor is optimized.

The invention discloses a tomato raw material loading, unloading and sorting method and system fused with AI visual recognition, and relates to the technical field of AI vision. A tomato raw material loading, unloading and sorting method fused with AI visual identification comprises the following steps that multiple frames of images are collected, qualified images are reserved through quality detection, the qualified frames are subjected to pixel-level segmentation, tomato segmentation masks are generated through multi-frame fusion, and blind areas are complemented; based on the complemented segmentation mask, high residual target points are judged, space attributes are extracted, the optimal moving path and action queue of the crane pipe are obtained, and a dynamic complementing and scanning queue is inserted to be issued and executed; and in the execution process, residual changes are monitored, the flushing effect is judged, dynamic recognition, complementary scanning and queue updating are conducted, parameters are adjusted, and sorting is conducted till discharging is completed. Accurate recognition, efficient path planning and dynamic supplementary scanning in the tomato loading, unloading and sorting process are achieved, and then the problems that in the prior art, due to insufficient recognition precision, path planning is unreasonable, and residual areas are missed in scanning are effectively solved.

The invention discloses a sports equipment management personnel identification method, system and equipment, and the method comprises the steps: starting and initializing a camera through detecting a sports equipment use trigger signal, and stopping automatic focusing; and shooting multiple frames of images at different focal lengths according to a preset interval, determining a clear area by using a gradient magnitude algorithm, and performing image registration and fusion to form a composite image. And if the whole area is not covered, shooting and splicing are carried out again. And through composite image splicing, distortion cutting, noise reduction and normalization processing, a final image is output for determining the identity of a person. According to the invention, through multi-frame shooting and accurate image processing, a high-quality image is obtained, and information limitation and quality defects of a single image are effectively avoided. Various algorithms are fused to ensure that the image is clear, complete and standardized, the accuracy and reliability of personnel identity recognition are remarkably improved, an accurate image basis is provided for sports equipment management, the normalization and safety of equipment use are guaranteed, meanwhile, the image collecting and processing efficiency is improved, and the management cost is reduced.

The invention provides a river surface flow velocity estimation method and device fusing multi-mode optical flow estimation and PINN, and relates to the technical field of hydrological monitoring, and the method comprises the steps: obtaining meteorological observation data and continuous multi-frame river surface images; performing multi-scale feature processing on the plurality of frames of river surface images to obtain a plurality of related pyramid feature data; carrying out coding processing on the meteorological observation data and at least one frame of image in the continuous multiple frames of river surface images to obtain context mixed feature data; and obtaining an optical flow field according to the multiple pieces of related pyramid feature data and the context mixed feature data, and correcting the optical flow field to obtain a target optical flow field. According to the scheme, high-precision prediction of the river surface flow velocity is realized.

The invention discloses an all-terrain maneuvering target tracking system and an all-terrain maneuvering target tracking method, belongs to the technical field of all-terrain moving target tracking, and aims to solve the problems of poor target tracking trajectory adaptability and large vibration interference in a complex terrain. During operation, the target orientation is positioned, the advancing direction is updated in real time, after the collection range is dynamically delimited, multiple frames of images are collected, and a topographic map is constructed through cutting, splicing and distortion correction; preprocessing the topographic map, extracting landform features, and dividing passable areas; generating a tracking trajectory in the passable area, and optimizing the trajectory through the position change of the target; and adaptively adjusting parameters of the balance component according to the track area, and fusing a vibration signal to judge whether to trigger track updating so as to ensure that the track dynamically adapts to terrain and target movement. According to the invention, through enhancing image processing and multi-source data fusion, the continuity, accuracy and environmental adaptability of target tracking in a complex terrain are significantly improved, and the reliability and efficiency of operation are effectively enhanced.

The invention discloses a multi-view athlete collision detection penalty auxiliary method for a Kabadi competition, which comprises the following steps: acquiring videos of different views of the Kabadi competition, respectively carrying out target detection, and generating a detection frame of an athlete; performing three-dimensional reconstruction on the position of the athlete according to the detection frame data of the multiple visual angles at the same moment, and analyzing to obtain space-time behavior data of the athlete; for each view angle, according to the athlete detection frame and the spatiotemporal behavior data, detecting whether potential collision occurs in the current frame, and if potential collision occurs in multiple continuous frames, marking the current frame as an event frame cluster; performing posture detection on the athletes in the event frame cluster, calculating the nearest distance between joint points of an attacking party and a defending party, selecting a frame with the minimum comprehensive distance as a penalty key frame, and marking the frame as a collision event; and analyzing and converting the space-time behavior data and the collision event of the athlete and a reconstructed image generated by a plug flow rear end into a live broadcast picture, state information and event prompt for a referee to understand and outputting the live broadcast picture, the state information and the event prompt.

A computerized technique is disclosed of identifying object features in an environment of a vehicle. The technique includes receiving, by an encoder, data representing a plurality of frames, the frames providing point-in-time versions of a segmented pointed cloud derived from output of one or more radar sensors of the vehicle and including points that represent radar detections corresponding to an object in the environment at respective instants in time. The technique further includes arranging the plurality of frames in a time-ordered queue and processing the frames in the queue, including (i) selecting, from among the points, a plurality of sample points that spans multiple frames of the queue, (ii) forming a plurality of groups of points based on respective sample points of the plurality of sample points, and (iii) extracting features of the object based on the plurality of sample points and the plurality of groups.

The invention discloses a display picture detection method and device, equipment, a medium and a product, relates to the technical field of image recognition, and utilizes an image classification model to analyze multiple frames of images so as to determine an initial classification result. And respectively extracting an edge feature map, a diagonal gradient feature map, an optical flow feature map and a difference feature map from the multiple frames of images. And based on the edge relative displacement between the edge feature maps of the adjacent frames and the gradient relative displacement between the diagonal gradient feature maps of the adjacent frames, determining a first identification result whether the display picture of the image flickers or not. And according to the initial classification result, the optical flow structure similarity corresponding to the multiple frames of optical flow feature maps and the difference structure similarity corresponding to the multiple frames of difference feature maps, determining a second identification result whether the display picture of the image is lagged or not. Different types of feature maps in multiple frames of images are analyzed, so that the abnormal problem of the display picture can be timely and accurately detected, and the maintainability of the system is improved.

The invention discloses a method and a system for generating a real-time dialogue image of a digital life entity, and relates to the technical field of artificial intelligence, the method comprises the following steps: acquiring audio and video information, the audio and video information comprising multiple frames of image information and audio information comprising a target person, and the image information and the audio information being in time synchronization; identifying image emotion information in the image information; performing semantic analysis on the audio information to generate semantic information; generating target semantic information according to the image emotion information and the semantic information; generating target prompt word information according to the target semantic information; and inputting the target cue word information into a pre-trained dialogue model to generate dialogue information of the digital life entity and the target person. The obtained target prompt word contains rich and accurate effective information, the current state of the target person can be accurately reflected, and the digital life body has high dialogue ability and is more real.

The invention discloses a cutting machine tool wear detection method and system based on image recognition, and relates to the technical field of wear detection, and the method comprises the steps: collecting multiple tool nose images in a short time window through triggering signal control, obtaining the time sequence information of a tool state, and eliminating the interference through combining with an image preprocessing and time sequence alignment technology. A U-Net model is used for precisely segmenting a single-frame image to obtain a tool abnormal area, dynamic association among multiple frames of images is deeply excavated through connected component analysis and time sequence consistency verification, a real abrasion area is screened out from a multi-frame sequence, and finally a tool abrasion detection result is obtained. Therefore, non-wear interference can be effectively distinguished and eliminated, and the defects that a traditional method is insufficient in precision and prone to false detection and missing detection are overcome, so that the accuracy and robustness of a detection result are remarkably improved, false alarm and missing alarm are avoided, and the intelligent production requirement of the modern industry is met.

The invention relates to the technical field of stereoscopic vision, and discloses a synchronous speed visual matching method and system, electronic equipment and a storage medium, and the method comprises the steps: synchronously collecting a stereoscopic video sequence with a predefined frame rate; executing multi-target hybrid tracking and motion induction detection, and outputting target motion information including position and velocity vectors; extracting hierarchical motion features of the target from continuous multiple frames of the stereoscopic video sequence, and performing unified space-time coding; under geometric constraints of stereoscopic vision, scale cosine similarity, direction similarity and trajectory consistency measurement are calculated and serve as observation evidences to be input into the probabilistic reasoning model for fusion, and a posterior probability representing matching reliability is output; the weight distribution of the speed similarity and the direction similarity is adjusted according to the motion characteristics of the targets in the scene, and the stable corresponding matching relation between the left view target and the right view target is established. According to the method, high-time-resolution information can be utilized, motion features and geometric constraints can be effectively fused, and the method has self-adaptive capacity.

The invention discloses a 4K multi-frame-rate YUV422 low-delay collaborative coding and decoding method, which relates to the technical field of video coding and decoding, and comprises the following steps of: configuring coding and decoding device parameters, establishing network connection, and setting a multi-frame-rate range and a buffer area; acquiring a 4K YUV422 original frame sequence from a video source and temporarily storing the 4K YUV422 original frame sequence; analyzing the difference and texture change between the current frame and the previous frame, and calculating the comprehensive content complexity; dynamically selecting an optimal target frame rate according to the complexity value and the network delay; the codec exchanges a frame rate switching command through a control channel and realizes synchronization, adjusts a compression parameter according to a selected frame rate, and generates a bit stream; and periodically exchanging network state information while transmitting the coded data. According to the invention, by introducing a dynamic frame rate collaborative adjustment mechanism based on content complexity, adaptive matching of coding parameters, scene contents and network states can be realized, and the problem that the stability of video image quality cannot be ensured while low delay is maintained in the prior art is solved.

The invention relates to the field of image processing, in particular to an oil cup defect detection method and system based on deep learning. The method comprises the following steps: acquiring continuous multi-frame oil cup images by using an industrial camera, preprocessing the continuous multi-frame oil cup images, and inputting the preprocessed oil cup images into a neural network model so as to obtain oil cup defect types; the defect types of the oil cup comprise bubbles, cracks, dirty points, scratches and normality; the preprocessing comprises the following steps of: performing distortion removal on an acquired oil cup image by using calibration parameters of an industrial camera, identifying a cup body of the oil cup, cutting the oil cup image, and only reserving a cup body area so as to obtain a first oil cup image; and carrying out multi-frame alignment on the first oil cup image corresponding to each frame of oil cup image, and carrying out synthesis processing on the aligned images so as to obtain a synthesized image. By adopting the method, the accuracy and the detection efficiency of the detection result of the electronic cigarette liquid cup can be effectively improved.

The embodiment of the invention provides a vehicle, a three-dimensional occupancy prediction method thereof and a training method of a bidirectional decoder, and the three-dimensional occupancy prediction method of the vehicle comprises the steps: sensing a three-dimensional space where the vehicle is located, and obtaining a multi-frame image, the multi-frame image comprising a current frame image and a preset number of historical frame images; generating a three-dimensional fusion feature based on the two-dimensional feature of the multi-frame image; performing feature interaction on the three-dimensional fusion feature, the initial query vector of the current frame and the initial query vector of the future frame by using a bidirectional decoder to obtain a target query vector of the current frame and a target query vector of the future frame; and performing three-dimensional reconstruction based on the target query vector of the current frame to generate a three-dimensional occupation prediction result of the current frame, and performing three-dimensional reconstruction based on the target query vector of the future frame to generate a three-dimensional occupation prediction result of the future frame. According to the invention, the technical problem of low accuracy of 3D occupancy prediction of the current frame by using the historical frame is solved.

A streaming neural network video codec that leverages temporal redundancy, processing video frames in a rolling window fashion. By encoding video frames using information from multiple frames and only transmitting essential codewords, the system ensures efficient compression with reduced computational overhead. Resiliency to lost codewords is achieved by training with random masks on one or more codewords so that the decoder is robust to packet losses. These techniques achieved improved compression efficiency and significantly reduces the average operations required per frame, allowing for real-time, high-quality video streaming.

The invention provides a difficult case recognition method, system and device which are used for mining possible difficult case scenes in the vehicle driving process and achieving accurate mining of the difficult case scenes. The method comprises the steps that firstly, a sensing data set collected in the vehicle driving process is obtained, the sensing data set is obtained by collecting a current scene through at least one sensor in a vehicle, and sensing data comprises multiple frames of sensing data; then the multi-frame sensing data is input into a sensing model to obtain a plurality of sensing results, and the sensing model is used for detecting elements in the input data and can output information of the detected elements, such as types, positions, shapes or sizes of the elements; then, according to the similarity among the multiple perception results, a difficulty coefficient is determined, and the difficulty coefficient is used for measuring the difficulty of determining a vehicle driving decision in the current scene or used for measuring the perception ability of the vehicle or the perception model in the current scene.

The invention relates to a traffic information determination method and device, electronic equipment and a vehicle, and relates to the technical field of data processing, the method comprises the following steps: obtaining a visual identification result and a map calibration result, the visual identification result is obtained by processing a multi-frame detection image comprising a target signboard through a target identification model, and the map calibration result is obtained by processing a multi-frame detection image comprising a target signboard; and according to the type of the target signboard and the environmental information, determining the confidence coefficient of the visual identification result, and then according to the visual identification result, the map calibration result and the confidence coefficient of the visual identification result, obtaining the target traffic information. The confidence of a visual identification result is determined through the type of the target signboard and the environment information, the visual identification result is obtained by processing multiple frames of detection images including the target signboard through the target identification model, the accuracy can be improved, and the accuracy of the visual identification result is improved based on the visual identification result, the map calibration result and the confidence of the visual identification result. And more accurate target traffic information can be obtained.

The invention relates to the technical field of welding tools, in particular to an aluminum alloy frame welding forming tool and process suitable for an exposed frame glass curtain wall, and the aluminum alloy frame welding forming tool comprises a clamping mechanism which is divided into an unfolding state for placing a plurality of frames and a clamping state for aligning the plurality of frames end to end; and the clamping mechanism comprises at least four groups of clamping parts which are annularly distributed, adjusting rods for pushing the clamping parts to clamp and an adjusting disc for driving the adjusting rods to horizontally move. Through the design of an elastic clamping structure and a folding elastic piece, rapid clamping and stable keeping of a single frame are completed, displacement or slippage during welding is avoided, meanwhile, a buffering and avoiding space is provided in the alignment process, synchronous horizontal displacement of multiple sets of adjusting rods is completed through a linkage mechanism of an adjusting disc and a V-shaped distance adjusting groove, and the welding precision is improved. And the clamping parts are pushed to be converted into a vertical clamping state from an unfolding state, so that the multiple frames are accurately aligned end to end at a time, and the technical effect of improving the welding precision and consistency is achieved.

The invention relates to a universal visual model training method and system, and belongs to the technical field of computer vision, and the method comprises the steps: obtaining a to-be-detected video sample, and extracting the multi-level and multi-scale features of continuous multi-frame images in a video sequence through a backbone network; predicting an optical flow field from the to-be-detected frame to the target frame through an optical flow prediction network based on the features of the adjacent frames; calculating a motion displacement truth value by using truth value frame information of the to-be-detected frame and the target frame, and supervising and constraining the optical flow field; performing distortion alignment on the adjacent frame features by using the optical flow field, and fusing the adjacent frame features with the to-be-detected frame features to generate enhanced features; based on the enhanced features, target classification and positioning are carried out through a target detection network; and constructing a total loss function in combination with the target detection loss and the optical flow supervision loss, performing end-to-end joint optimization training on the backbone network, the optical flow prediction network and the target detection network, and outputting a video monitoring result by using the trained visual model.

This application provides a foldable electronic device for the field of communication technology. The foldable electronic device includes an antenna, a hinge assembly, a first housing, and a second housing. At least one of the first and second housings is rotatably connected to the hinge assembly, and the angle between the first and second housings can be maintained at a preset angle. The first housing includes multiple frames, including a second frame and a third frame located at both ends of the hinge assembly, and a first frame connected to the ends of the second and third frames away from the hinge assembly. The antenna is located on one frame of the first housing and includes a main radiator and a feed point located on the main radiator. When the antenna is located on the first frame, the feed point is 1 / 8 to 3 / 8 of the antenna's operating wavelength from the edge of the ground facing the second or third frame. When the antenna is located on the second or third frame, the feed point is 1 / 8 to 3 / 8 of the antenna's operating wavelength from the hinge assembly. The foldable electronic device provided by this application enables satellite communication.

The invention provides a low-light environment imaging enhancement method based on multi-frame synthesis, and the method comprises the steps: carrying out the independent geometric correction of each region according to the optimal registration transformation matrix of each region, and obtaining the image data of each region after geometric correction; weight distribution is carried out on the high-frequency texture components according to the texture consistency evaluation result, if the texture direction deviation angle is smaller than a preset angle threshold value, the weight coefficient of the frame is improved, the high-frequency components of the multiple frames are fused, and a high-frequency synthesis component with enhanced texture details is obtained; reconstructing the texture detail enhanced high-frequency synthetic component and the contour retentivity optimized low-frequency component, and performing multi-scale fusion processing on the reconstructed image to obtain a final nighttime animal observation synthetic image; in an image quality evaluation module of the camera APP, animal key feature points are extracted from the synthesized image, the reliability of animal feature recognition is verified by calculating the stability of feature point descriptors, and a verification result of the reliability of animal feature recognition is obtained.

The invention relates to the technical field of computer vision, particularly provides a training method of a video time positioning model, a video time positioning method, equipment and a medium, and aims to solve the problem of large video time positioning error. In order to achieve the purpose, the model training method comprises the steps that multiple frames of images are sampled from a training video to serve as training data, the training data and a first preset query text are coded to obtain visual features and text query features, and a video time positioning model is trained based on the visual features and the text query features, obtaining a plurality of candidate answers, respectively calculating the relative advantage value of each candidate answer based on the real answer of the first preset query text and the plurality of candidate answers, adjusting the parameters of the video time positioning model based on each relative advantage value, and continuing to execute the step of sampling multiple frames of images from the training video. Therefore, the performance and accuracy of video time positioning can be improved.

The embodiment of the invention provides a target tracking method. The method comprises the following steps: converting a video into a plurality of frames of images; obtaining a target box prompt file; constructing an efficient feature extraction module and a lightweight feature extraction module, performing frame-by-frame reasoning on the multiple frames of images to obtain a target area, and updating the target box prompt file; and according to the target frame prompt file, drawing a frame in a target area of the image. According to the target tracking method provided by the embodiment of the invention, by combining the target box prompt file and frame-by-frame reasoning, the position and motion information of the target can be more accurately captured, and the conditions of tracking loss and inaccurate tracking are reduced. The embodiment of the invention further provides a target tracking device and electronic equipment.

The invention discloses a tennis serving recognition method and device based on a time interval, and the method comprises the steps: collecting video frames in parallel through a plurality of cameras, and carrying out the time alignment of the video frames, thereby obtaining a plurality of synchronous frames; recognizing two-dimensional pixel coordinates of the tennis ball in the multiple paths of synchronous frames, and constructing three-dimensional coordinates of the tennis ball in the space through a ray intersection method; analyzing the continuous three-dimensional coordinates of the tennis ball in the sliding window in the space, and judging the movement trend of the tennis ball on the y axis; if the motion trend on the y axis is continuous rising or falling, calculating the frame interval between the current frame and the last effective motion; finally, judging whether the frame interval is greater than an effective motion interval threshold value or not; if yes, new serving is judged, and the coordinates of the serving starting point are recorded. According to the method, the y-axis change trend of the tennis movement is analyzed through the multi-frame sliding window, the serving movement characteristics are accurately recognized in combination with the frame interval threshold value, serving and common hitting can be accurately distinguished according to the serving movement characteristics, and the misjudgment rate is effectively reduced.