35 results about "Behavior space" patented technology

The invention discloses a space-time behavior detection method, which comprises the following steps: carrying out object detection on all frames in a sample video to obtain a candidate object set; calculating all inter-frame optical flow information in the sample video, and obtaining a motion set; constructing a space-time convolution-deconvolution network for constructing additional object attention mechanism and motion attention mechanism; after performing space-time convolution processing on each time slice of the sample video, adding corresponding sparse variables and sparse constraints toobtain a network structure S; training the network structure S by taking the classification loss based on the cross entropy and the loss of the sparse constraint as objective functions; and calculating behavior types and sparse coefficients corresponding to the time slices in the test sample video, and obtaining an object behavior space-time position. According to the method, through object detection and optical flow prediction, the behavior search space is reduced, and space-time behavior detection has good robustness.

The invention discloses an integrated monitoring system and an integrated monitoring method for the seepage behavior of water engineering under complex environments. The integrated monitoring system comprises a seepage behavior space-time monitoring device and a sensor fiber seepage-detecting sensibilizing device, wherein the seepage behavior space-time monitoring device comprises a vertical force bearing fiber pillar, an outer edge through pipe and a sensor fiber; a left force-bearing beam and a right force-bearing beam are respectively arranged at the two sides of the vertical force-bearing fiber pillar; the outer edge through pipe is arranged on the vertical force-bearing fiber pillar in a sleeving manner; a fiber collecting box is arranged above a second transition round end; the sensor fiber in the fiber collecting box passes through the outer edge through pipe and is connected with an element bearing carrier containing a temperature measuring device, and then passes by the second transition round end and a first transition round end in turn, penetrates through an elastic device and is led from a third transition round end. According to the integrated monitoring system for the seepage behavior of the water engineering under complex environments provided by the invention, the research and development for basic sensor fiber and serial products and techniques, such as secondary processing for common sensor fiber, etc., are put forward, and the space-time horizontal and longitudinal quantitative and qualitative evaluation can be realized.

A legged mobile robot the body of which can be controlled in attitude stabilization by using ZMP stability determining norms with a relatively slow sampling period. The legged mobile robot has a ZMP stable region comprising a supporting polygon formed by a sole contact point of a mobile leg with the road surface, and a ZMP behavior space in which deformation or momentum of the robot body is generated so that the ZMP is directed toward generally the center of the ZMP stable region. Regarding the sign of the deformation (or momentum) of the robot in the ZMP behavior space, the negative sign is defined to be the direction in which a spatial distortion causing the ZMP toward the edge of the stable region is produced while the positive sign is defined to be the direction in which a spatial distortion causing the ZMP toward the center of the stable region is produced.

The invention discloses a humanoid robot stable control method of an RBF-Q learning frame. The method comprises the following steps: the RBF-Q learning frame which solves the problems of state space serialization and behavior space serialization in a Q learning process is brought forward; an online motion adjusting stable control algorithm of the RBF-Q learning is brought forward, loci of the hip joint, the knee joint and the ankle joint of a support leg are generated, and a humanoid robot is controlled to walk stably through calculation of angles of other joints; and finally, the feasibility and validity of an RBF-Q learning frame method are verified on the Vitruvian Man humanoid robot platform designed by the laboratory. The method provided by the invention can generate a stable walking gait of the humanoid robot in an online learning process.

The invention relates to an intelligent decision-making method for data packet transmission based on deep reinforcement learning, including: constructing a deep neural network model; designing and initializing state space and behavior space; obtaining current state information and historical state information of data transmission, and inputting state space; adopting The experience playback mechanism saves historical state information; iteratively executes steps (3) and (4) T times, and the round ends; updates the target value neural network parameter θ′, and assigns the latest parameter θ of the original value neural network to the target value neural network ; Iteratively execute steps (2) to (5) until the number of iterations reaches the preset round upper limit N or the deep neural network converges, then terminate and automatically obtain a data transmission strategy that satisfies multiple constraints and lower energy consumption. The invention improves the service quality of users, reduces the energy consumption of data transmission at the same time, and effectively improves the intelligent decision-making ability of data transmission in a communication network in a highly complex and dynamic environment.

The invention relates to a human behavior identification method based on multi-feature fusion, which comprises the following steps of: acquiring a human body behavior video by using a camera, extracting a foreground image of each frame of image, and carrying out cavity filling and interference filtering to obtain a human body silhouette image sequence; Calculating the similarity between adjacent frames in the image sequence, and obtaining the weight of each frame of image representing the behavior posture; According to each frame of image in the human body silhouette image sequence and the corresponding weight thereof, obtaining an action energy diagram representing a behavior process through weighted average; And extracting a Zernike moment, a gray histogram and a texture feature of the action energy diagram to form a multi-dimensional feature fusion vector containing behavior space-time characteristics; Constructing feature vector template libraries of different standard behaviors; And in the behavior identification process, extracting feature vectors of to-be-identified behaviors according to the to-be-identified videos, matching the feature vectors of the to-be-identified behaviors with the feature vectors of the standard behavior template library one by one, determining behavior types according to matching results, and realizing accurate identification of human body behaviors. According to the method, the time change and spatial attitude characteristics of human body behaviors are represented by constructing the action energy diagram, the behavior recognition accuracyand real-time performance can be improved, and certain practical value is achieved.

The invention discloses a microcosmic decision-making method for an automatic driving vehicle based on reinforcement learning. The method adopts the A3C algorithm of reinforcement learning, the driving behavior is output by the Actor network, the flexibility is strong, and the complexity of the judgment logic is not affected by the size of the state space and the behavior space. The method employs a two-stage training solution process. In the first stage of training, a microcosmic decision-making model for automatic driving is obtained for all road sections to ensure driving safety. In the second stage, the overall model of the first stage is deployed to each road segment, and each road segment trains a single-segment model on this basis, which is portable. Meanwhile, the continuous training in the second stage enables the method to adapt to the influence of various real-time factors. Finally, the distributed communication architecture based on the real Internet of Vehicles system structure is described, which can complete the distributed calculation in the solution process. Therefore, the method can adapt to different road characteristics and dynamic driving environments, and has wide applicability and robustness. sex.

The invention discloses a complex behavior recognition method based on video basic unit analysis. This method splits video behavior recognition into temporal atomic behaviors, spatially extracts object and background information in the video based on target detection and scene recognition, and sends the extracted semantic information to a temporal model for analysis. Compared with the previous video behavior recognition, this method decomposes the video into basic tasks in time and space for detection and recognition, which is more interpretable. At the same time, based on this method, the required basic unit information can be selectively extracted for different task situations, and the flexibility of complex behavior recognition tasks is enhanced through the method of splitting.

The present invention proposes a control method for mooring assisted dynamic positioning system based on reinforcement learning, including: first constructing a Markov decision model of the optimal point selection problem, constructing a state space and a behavior space; using a neural network to construct a reinforcement learning model Q function, Based on the current state of the mooring-assisted dynamic positioning system measured in real time, the control system uses the ε-greedy algorithm to select behaviors, and observes the system state s' and feedback rewards after the behavior a is selected; the state, behavior, obtained rewards and The new state is stored in the memory bank as label data, and the neural network is trained; by repeating the above process, the mooring-assisted dynamic positioning system can obtain the behavior selection strategy that maximizes the reward function, and obtains the power consumption of the pusher under the anchor point control mode. The lowest sweet spot.