1168 results about "Multi target tracking" patented technology

A collision sense and avoidance system and method and an aircraft, such as an Unmanned Air Vehicle (UAV) and / or Remotely Piloted Vehicle (RPV), including the collision sense and avoidance system. The collision sense and avoidance system includes an image interrogator identifies potential collision threats to the aircraft and provides maneuvers to avoid any identified threat. Motion sensors (e.g., imaging and / or infrared sensors) provide image frames of the surroundings to a clutter suppression and target detection unit that detects local targets moving in the frames. A Line Of Sight (LOS), multi-target tracking unit, tracks detected local targets and maintains a track history in LOS coordinates for each detected local target. A threat assessment unit determines whether any tracked local target poses a collision threat. An avoidance maneuver unit provides flight control and guidance with a maneuver to avoid any identified said collision threat.

The invention provides a video monitoring system, which comprises a first image pick-up system, a second image pick-up system and a control system. The first image pick-up system comprises one or more cameras for shooting a wide-angle video within a large-scene vision field; the second image pick-up system comprises one or more pan / tilt / zoom (PTZ) cameras for shooting a local video within the large-scene vision field; and the control system comprises an image acquisition module, a foreground extraction module and a coordinate conversion module. The image acquisition module is used for receiving the wide-angle video shot by the first image pick-up system; the foreground extraction module is used for extracting a target of interest from the wide-angle video; and the coordinate conversion module is used for converting a two-dimensional coordinate (x, y) of the projection of an arbitrary point shot by the first image pick-up system in the large-scene vision field on an image plane of the wide-angle video shot by the first image pick-up system into a vertical altitude angle theta and a horizontal azimuth angle phi when one selected PTZ camera in the second image pick-up system is aligned with the arbitrary point in a picture center by a coordinate conversion mechanism, wherein the coordinate conversion mechanism is established by at least three arbitrary points selected randomly in the large-scene vision field.

The invention discloses a cross-camera pedestrian detection tracking method based on depth learning, which comprises the steps of: by training a pedestrian detection network, carrying out pedestrian detection on an input monitoring video sequence; initializing tracking targets by a target box obtained by pedestrian detection, extracting shallow layer features and deep layer features of a region corresponding to a candidate box in the pedestrian detection network, and implementing tracking; when the targets disappear, carrying out pedestrian re-identification which comprises the process of: after target disappearance information is obtained, finding images with the highest matching degrees with the disappearing targets from candidate images obtained by the pedestrian detection network and continuously tracking; and when tracking is ended, outputting motion tracks of the pedestrian targets under multiple cameras. The features extracted by the method can overcome influence of illuminationvariations and viewing angle variations; moreover, for both the tracking and pedestrian re-identification parts, the features are extracted from the pedestrian detection network; pedestrian detection, multi-target tracking and pedestrian re-identification are organically fused; and accurate cross-camera pedestrian detection and tracking in a large-range scene are implemented.

The invention discloses an intelligent multi-target active tracking monitoring method and system. The method comprises a plurality of steps such as a multi-target active tracking step, an active focusing step, a target switching step, a relay tracking step, an access warning step and a complicated environment preprocessing step. A panoramic picture in a monitor region is acquired by virtue of a first camera, a tracked target is locked, real-time coordinate information of the tracked target in the panoramic picture as well as a first angle control value of a cradle head of a second camera corresponding to the coordinate information and a first focusing value of the second camera are calculated, when a close-up image of one tracked target needs to be acquired, the second camera positions the tracked target according to the angle control value and the first focusing value corresponding to the selected tracked target so as to acquire the close-up image, so that the second camera is always focused on the tracked target to continuously track and photograph the tracked target in real time. By adopting the intelligent multi-target active tracking monitoring method and system, the cradle head control problem and focusing problem in the video monitoring and multi-target tracking can be effectively solved.

A system for applying video data to a neural network (NN) for online multi-class multi-object tracking includes a computer programed to perform an image classification method including the operations of receiving a video sequence; detecting candidate objects in each of a previous and a current video frame; transforming the previous and current video frames into a temporal difference input image; applying the temporal difference input image to a pre-trained neural network (NN) (or deep convolutional network) comprising an ordered sequence of layers; and based on a classification value received by the neural network, associating a pair of detected candidate objects in the previous and current frames as belonging to one of matching objects and different objects.

The invention discloses a KCF and Kalman-based improved multi-target tracking method comprising the following steps: using a GoogLeNet network model to detect targets and extract characteristic vectors of the targets; combining the prediction position of each target on the current frame in a previous frame tracking link with the current frame target observation position, the overlapping rate and the characteristic vector spatial distance, thus building an association matrix, and using a matching algorithm for matching; updating the tracking box of the directly matched tracking link and the corresponding characteristic vector; using a KCF tracker to locally track a target failed in matching; carrying out weighted fusion for the KCF tracking result and a Kalman tracking result, thus obtaining a position, and updating same; predicting the next frame position of each target in the tracking link. The method can combine with the CNN network to extract the characteristic vectors, and uses theKCF local tracking to improve the tracking effect, thus well solving the target blocking and target error detection problems; in addition, the invention also provides a multi-target tracking system and device.

The multi-target tracking and discrimination system (MOST) fuses with and augments existing BMDS sensor systems. Integrated devices include early warning radars, X-band radars, Lidar, DSP, and MOST which coordinates all the data received from all sources through a command center and deploys the GBI for successful interception of an object detected anywhere in space, for example, warheads. The MOST system integrates the optics for rapid detection and with the optical sensor array delivers high-speed, high accuracy positional information to radar systems and also identifies decoys. MOST incorporates space situational awareness, aero-optics, adaptive optics, and Lidar technologies. The components include telescopes or other optical systems, focal plane arrays including high-speed wavefront sensors or other focal plane detector arrays, wavefront sensor technology developed to mitigate aero-optic effects, distributed network of optical sensors, high-accuracy positional metrics, data fusion, and tracking mounts. Field applications include space monitoring, battlefield artillery, battlefield management, ground defense, air defense, space protection, missile defense, gunfire detection, and the like.

The invention discloses a multi-target tracking method based on depth track prediction. The method comprises the following steps: constructing a track prediction model based on a long-short time memory network for a multi-target tracking system; using the trajectory data of the real tracking scene to train a trajectory prediction model; constructing conservative short-time trajectory fragments byusing the appearance characteristics of target detection, and calculating the appearance similarity among the trajectory fragments; carrying out depth track prediction on the target on line by using the trained track prediction model, obtaining the motion similarity between track segments, comprehensively considering the appearance similarity and the motion similarity, and setting a network modelof target tracking to complete multi-target tracking. According to the method, a long-short time memory network-based trajectory prediction model is constructed for a multi-target tracking system, andcompared with a traditional method, the method can fully consider the historical trajectory information and scene information of the target, calculate the inter-target motion similarity with better robustness, and further improve the multi-target tracking effect.

The invention relates to a large-scene multi-target tracking shooting video monitoring system. The large-scene multi-target tracking shooting video monitoring system comprises a large-scene camera shooting sub-system and a target tracking camera shooting sub-system, wherein the large-scene camera shooting sub-system adopts a high-point fixed focus camera used for shooting a video in a large-scene vision field; the target tracking camera shooting sub-system adopts one or more PTZ (Pan / Tilt / Zoom) cameras used for shooting a local video of a moving target in the large-scene vision field in a tracking way; the output end of the large-scene camera shooting sub-system is connected with the input end of the target tracking camera shooting sub-system. The invention further discloses a large-scene multi-target tracking shooting video monitoring method. According to the system and the method, accurate interaction and control between the large-scene camera shooting sub-system and the target tracking camera shooting sub-system are effectively achieved, the corresponding PTZ camera is designated according to a preset priority sequence for movement shooting, a tracked target is located in the imaging center of each PTZ camera all the time, the complexity of the system is reduced, and the image quality and the target tracking reliability of the monitoring system are greatly improved.

The embodiment of the invention provides a multi-target tracking method and system based on depth features. The method comprises the following steps: obtaining detection frame positions correspondingto targets detected in a current frame image and the depth features of the targets; based on the position of the detection frame corresponding to each target in the previous frame of image, obtainingthe prediction position of each target in the current frame by using a Kalman filter; according to the detection frame position corresponding to each target, the prediction position of each target inthe current frame, the depth feature of each target and the depth feature set of each tracker, performing cascade matching on the detection frame corresponding to each target and the tracker by usinga Hungarian algorithm; And calculating an IOU distance matrix between the detection frame on the non-cascade matching and the tracker to be matched, and performing IOU matching between the detection frame and the tracker by using a Hungarian algorithm based on the IOU distance matrix to obtain a final matching set. According to the embodiment of the invention, the target tracking effect under theshielding condition can be effectively improved, and the number of times of ID switching is reduced.

The invention provides a video flame detecting method based on a multi-feature fusion technology. The video flame detecting method includes firstly using a cumulative geometrical independent component analysis (C-GICA) method to capture a moving target in combination with a flame color decision model, tracking moving targets in current and historical frames in combination with a multi-target tracking technology based on moving target areas, extracting color features, edge features, circularity degrees and textural features of the targets, inputting the features into a back propagation (BP) neural network, and further detecting flames after the decision of the BP neural network. According to the video flame detecting method, spatial-temporal features of the moving features, color features, textural features and the like are comprehensively applied, the defects of algorithms of existing video flame detecting technologies are overcome, and reliability and applicability of the video flame detecting method are effectively improved.

The invention discloses a monitoring video multi-target tracking method based on deep learning. The method includes the steps of firstly encoding a video, extracting an image sequence from the encoded video, then preprocessing and inputting images into a trained Faster R-CNN network model, the Faster R-CNN network model extracting target location information and target spatial features from corresponding layers of the network; inputting the target location information and the target spatial features into an LSTM network, and predicting locations of targets at a next moment; and subjecting the spatial features of the targets to a fusion method to obtain fusion features of the targets at the next moment, adding different weights through the location similarity and spatial feature similarity to obtain the final similarity, and then judging the corresponding relationship of the multiple targets detected at the current moment and the multiple targets in a tracking state at the previous moment. By target tracking, the method of the invention can reduce the missed detection rate of multi-target tracking, improve the accuracy of multi-target tracking, and solve the target occlusion problem in a short time in the tracking process.

An automatic identification system, which is integrated by the information of a multi-visual sensor and is used on inland river ships, comprises a large-scale monitoring visual sensor which can be used on monitor fairways, an express-ball visual sensor which can be used for shooting close-up images of ship bodies and name plates of ships, and a microprocessor which can be used for tracing target ships, identifying images and summarizing traffic situations of inland rivers; the large-scale monitoring visual sensor can realize multi-target tracing for ships on fairways; when a ship enters the monitoring area, the system automatically produces an event and an ID of the ship to control the rotating and focusing of the express-ball visual sensor; the express-ball visual sensor focuses on the traced vessel to shoot; the height of the vessel body which is above water and the load can be estimated by detecting of the outline of the close-up image; and at the same time, by positioning the ship cockpit, the number of the name plate can be shot and identified; by integrating multi-visual sensors and automatically collecting the basic data of inland river traffic by a computer, the present invention can effectively manage inland rivers.

The invention relates to a multi-target tracking system based on deep learning and an implementation method. The method comprises the following steps: getting the target position of a first frame through target detection, and adding multiple to-be-tracked targets to a tracking queue; inputting a next frame of image and traversing the tracking queue to get the position of the target in the next frame; after getting the position of the target in the next frame, judging whether the target is off the screen based on thresholds; if the target is not off the screen, invoking target detection every other fixed frame, and calculating the IOU (intersection over union) of the target detection result and the tracking result; if IOU<0.1, judging that a new target is added to the screen, and adding the target to the tracking queue; if IOU>0.5, replacing the tracking box with the target detection box to correct the position; and continuing target tracking. By carefully designing the network structure and improving the training method, under the condition of high tracking precision, the tracking speed is increased significantly, network redundancy is reduced, and the size of the model is reduced.

Provided are a sensor network structure, a data aggregation method, and a clustering method for efficient multi-target tracking. The multi-target tracking may be efficiently performed in a heterogeneous sensor network by combining clustering methods and adaptively varying the clustering methods. As such, an energy consumption problem in a sensor network may be reduced, and a data transmission delay problem or a data traffic problem may be solved by reducing the amount of data to be transmitted.

The present disclosure provides a multiple targets-tracking method and apparatus, a device and a storage medium, wherein the method comprises: obtaining a to-be-processed current image, inputting the current image into a convolutional neural network model obtained by pre-training, and obtaining a target detection result; respectively extracting feature vectors of each detected target from a pre-selected convolutional layer; respectively calculating a similarity between the feature vectors of each target in the current image and feature vectors of each target of previous images, completing association of the same target between different image frames according to calculation results, and allocating a tracking serial number. The solution of the present disclosure may be applied to meet requirements for real-time processing.

The invention discloses a multi-target active recognition tracking and monitoring method. The method comprises the following steps of collecting a panoramic picture in a monitor area through a first camera, and after tracked targets are locked in the panoramic picture, computing real-time coordinate information of each tracked target in the panoramic picture, and a first angle control value of a tripod head of a second camera and a first focus value of the second camera which are corresponding to the coordinate information; and when a close-up picture of one tracked target needs to be collected, locating the close-up picture by the second camera according to the first angle control value and the first focus value which are corresponding to the selected tracked target at present, and collecting the close-up picture to enable the second camera to always focus on the tracked target to carry out real-time continuous tracking shot on the tracked target. The method effectively solves the problems of tripod head control and focusing in video monitoring and multi-target tracking.

The invention discloses a multi-target tracking method based on millimeter-wave radar. The method comprises acquiring point cloud data by millimeter-wave radar; clustering the point cloud data to distinguish the echo signals of different targets; and estimating the state information of multiple observed targets according to a clustering result so as to track the multiple targets. The method improves a DBSCAN clustering algorithm according to the characteristics of the millimeter-wave radar, improves the accuracy of esimating the number of the targets and the states of the targets, and predictsand tracks the trajectories of the multiple targets in a complex environment by using a Kalman filter and data association algorithm.

The invention relates to a method for implementing realistic game, which comprises the following steps of: (1) establishing a human body skeleton model; (2) establishing a game movement library under an offline state, establishing movement libraries respectively according to game items, and performing multi-frame movement decomposition on a single semantic movement;(3) calibrating a binocular camera to acquire parameters of the binocular camera and polar calibration; (5) background modeling; (6) selecting an interactive characteristic label; (7) foreground partitioning; (8) initializing the information of the characteristic label; (9) detecting human face and skin color; (10) multi-target tracking; (11) completing sparse stereo matching; (12) acquiring a 3D skeleton; and (13) matching with the movements in the offline movement libraries to realize movement recognition, combining single-frame image analytic matching with multi-frame image analytic matching to obtain the semantic movement, transferring the semantic movement to a game executing unit to implement the function of realistic game. Being stronger in interactivity and reality as well as simpler and more convenient in operation, the method of the invention is a game implementation method with low cost and is more suitable for being extensively accepted by general people.

The invention discloses a video structured processing method based on target behavior attributes. The method comprises the steps that the basic attributes of the target are acquired by using a YOLO target detection algorithm; the trajectory information of the detected target is acquired by using a multi-target tracking algorithm; abnormal video frames are extracted by using an abnormal behavior analysis algorithm based on the motion light flow characteristics; the corresponding target category attributes and the target trajectories and other characteristic information are acquired by using themethod according to the meta-data structure constructed by customization; the false detection data existing in the extracted meta-data are corrected by using a weighted judgment method; and the acquired data are uploaded to the rear-end server to be further processed. With application of the mode, the unstructured video data can be converted into the structured data having practical value so thatthe network transmission efficiency of the video monitoring system can be enhanced and the load rate of the rear-end server can be reduced. The invention also provides a real-time processing system based on the target behavior attributes and a real-time processing device based on the target behavior attributes.

The invention discloses a probability hypothesis density multi-target tracking method based on a variational Bayesian approximation technology, and belongs to the technical field of guidance and intelligent information processing. The probability hypothesis density multi-target tracking method based on the variational Bayesian approximation technology mainly solves the problem that an existing random set filtering method can not achieved varied number multi-target tracking under an unknown quantity measurement noise environment. According to the method, the variational Bayesian approximation technology is introduced, posterior probability hypothesis density of target states and measurement noise covariance is estimated in a combination mode, a Gaussian mixture inverse gamma distribution recurrence closed solution is adopted, and thus the varied number multi-target tracking under the unknown quantity measurement noise environment is achieved. The probability hypothesis density multi-target tracking method based on the variational Bayesian has a good tracking effect and robustness, is capable of meeting the design demands on practical engineering systems and has good engineering application value.

The invention discloses a traffic video intelligent analysis method based on target detection and tracking. The method specifically comprises the following steps: using highway videos acquired by using monitoring cameras; detecting vehicle target detection by using a deep learning method, acquiring target tracks by using a multi-target tracking method according to a detection result, automaticallyanalyzing the target tracks acquiring traffic flow and vehicle speed, detecting traffic congestion and parking traffic abnormal events and completing intelligent analysis of traffic videos. Vehiclescan be detected and tracked for a long time in a visual field range, so that traffic parameters are accurately obtained, and traffic events are detected. The invention provided by the invention has high stability when being used in various traffic scenes, and has certain practical value and wide market potential.

The invention discloses a three-dimensional multi-target tracking method fusing images and laser point clouds. The method is characterized by fusing the point cloud of a laser radar and the image dataof a camera to give full play to the complementary advantages between the point cloud data and the image data, extracting the three-dimensional space position information, the point cloud features and the image features of a target, matching a detection target and a tracking track, carrying out the state estimation on the tracking track in combination with a Kalman filter, thereby realizing the accurate and stable three-dimensional multi-target tracking. The method can be used for tracking and predicting the moving targets, such as pedestrians, vehicles, etc., in various unmanned vehicles, and can also be used in the fields of security monitoring, unmanned aerial vehicle-to-ground target reconnaissance and the like.

The invention discloses a method for generating video abstract on the basis of a deep learning technology. The method includes modeling backgrounds of video stream frame by frame and acquiring moving foregrounds to be used as candidate moving objects; tracking the candidate moving objects of each frame by the aid of a multi-object tracking algorithm and updating candidate objects which form movement tracks; training object classifiers by the aid of convolutional neural networks, confirming the candidate objects and determining categories of the objects by the aid of the classifiers after real moving objects are confirmed; fitting all the real moving objects and relevant information on a small quantity of images, forming video snapshots and displaying the video snapshots to users. The method has the advantages that the real objects and noise can be accurately differentiated from one another by the aid of the deep learning technology; the objects do not need to be confirmed frame by frame owing to an accurate multi-object tracking technology, accordingly, the computational complexity can be greatly reduced, an omission factor of the objects and a false alarm rate of the noise can be effectively reduced, the video processing speeds can be increased, and the method can be applied to various complicated scenes.

The invention discloses an invasion detecting alarming system based on multi-camera data combination and a detecting method thereof, and belongs to the technical field of video monitoring. The system comprises a common camera, a cloud camera, a zooming wide-angle multi-target tracking system, a PC type hard disk recorder, a control system, a display, a multi-stage alarming processing module and an invader recognizing and treating module. According to the system, an area to be monitored is completely covered with a plurality of cameras; the shot video signals can be digitized and compressed and then transmitted to the control system to recognize an invader and respond to the corresponding alarming level. The system is applicable to scenes or channels with high requirement on safety production, such as a factory area, a tunnel, a warehouse and an airport; the system can perform monitoring and alarming at different conditions.

The invention discloses a multi-target track method based on moving target detection in video monitoring. The method comprises the following steps: firstly, using a background removing method to detect foreground moving targets; then establishing an incidence matrix between a foreground target block mass of the current frame and the target detected in a previous frame so as to judge various states of the targets (such as target missing, target keeping initial condition, target overlap, target separation and the like), and secondarily tracking the targets which are in the separation condition; and at last, updating the positions and areas of the targets, kernel weighted color histograms and other characteristics to achieve tracking multiple targets. The method of the invention can be used for effectively and reliably tracking sheltered targets and the multiple targets which are mutually overlapped and then separated in real time, and improving stability and robustness of a video monitoring system.

The invention relates to a multiple trace point-based human body action recognition method. The method comprises the following steps of: based on an action requirement needing to be judged, setting at least one trace point on a human body or a sports apparatus to be tested; acquiring the space positions of each track point at different moments and recording the acquired space positions as a group of data points corresponding to the trace points; calculating corresponding action data of each trace point based on the action requirement needing to be judged by using the space position data of the group of data points corresponding to each trace point; and recognizing the movement action of the human body to be tested according to the corresponding action data of each trace point. The method can also recognize a human body gesture. The method has the advantages of realizing tracking of a plurality of targets, tracking a plurality of positions of the human body to be tested, recording the movement locus of a tracked position, positioning and describing the posture of the human body and truly reflecting the movement situation of the human body.

The invention discloses a multi-target tracking method based on a recurrent neural network, comprising the following steps: building a monitoring video data set marked with each frame of pedestrian location; manually expanding the monitoring video data set marked with each frame of pedestrian location to get training set samples; grouping the training set samples to get multiple training groups; building a multi-target tracking network; inputting the training groups in sequences to the multi-target tracking network for training; and inputting video data to be tested to the trained multi-target tracking network and performing forward propagation to get the moving trajectories of multiple targets. According to the invention, a proposed network model is trained in an end-to-end manner using raw data and a large amount of data obtained through artificial expansion, complex tasks such as data association and track estimation are completed under a unified neural network architecture, and the moving trajectories of targets can be tracked effectively under different directions, light conditions, deformations and other complex environments.