31results about How to "Reduce the number of data transfers" patented technology

The invention discloses an image processing method and device based on an embedded GPU and the convolution calculation. The method comprises the steps of optimizing the convolutional calculation in an SSD algorithm; performing matrix transformation on an input image by adopting the memory optimized convolution expansion; using a CUDA to parallelly process to form an intermediate matrix, meanwhile, adopting the convolution kernel matrix row and column expansion alignment, partitioning after the convolution kernel matrix expansion so as to reduce the memory overhead during operation, finally, adopting a highly optimized cuBLAS matrix multiplication function in a CUDA library to carry out convolution calculation parallel acceleration, and finally, combining and outputting the matrixes. According to the method provided by the invention, the memory overhead can be reduced, the performance of the algorithm is improved, the advantages of the GPU parallel control are brought into play, the matrix multiplication time is reduced, and the calculation efficiency is improved.

Provided is a motion estimation device in which an amount of pixel data transferred from an external frame memory to an internal reference local memory is reduced. By the motion estimation device, it is possible to reduce a memory capacity and a size or processing of a circuit controlling the pixel transfer. In a reference memory control unit and an internal reference memory, a height of a area to be updated is set to L pixels, where L is power of 2, a logical address segments, whose size is suitable for address calculation, are allocated to picture space, and FIFO management is performs. In another application, an assistance memory is added, and another element other than the assistance memory performs the FIFO management for rectangular areas in an image of a conventional width. As a result, the address calculation is simplified, which makes it possible to reduce an embedded circuit for the reference memory control unit and the internal reference memory.

The invention discloses a wireless network multi-path routing network transmission method aiming to solve the problem that the existing network coding transmission methods need many time slots, and have too many data transmission times and too high demand for nodes. The main steps of the method are network initialization and data transmission; the network initialization is mainly to determine thelevel of each node and the neighbor node information and complete networking; the data transmission is that a node judges the zone bits of message packets to perform different treatments to differentkinds of message packets, judges the zone bits of the levels of message packets to selectively receive the desired information, judges the local information pool and neighbor information list to select the optimal network coding and performs primary data transmission to complete information transmission and the information transmission of the node. The invention has the advantage of less time slots and data transmission times and lower demand of nodes for network and can be used for multi-user wireless data transmission.

The invention discloses a reconfigurable unit supporting multiple working modes and the working modes thereof. The reconfigurable unit is characterized by comprising a control layer, an arithmetic layer and a storage layer; the control layer comprises a state layer interface, a configuration layer interface, a data layer interface, an address generator and a controller; the arithmetic layer comprises an arithmetic device; the storage layer comprises a source operand cache unit and a destination operand cache unit. The working modes of the reconfigurable arithmetic unit comprise the storage arithmetic mode, the pulse arithmetic mode and the stream arithmetic mode, and higher flexibility is provided for algorithm mapping of a computing system. When task mapping is carried out in the computing system, the specific working modes of the reconfigurable arithmetic unit can be selected according to specific features and the bottleneck of the algorithm to be mapped and in combination with specific conditions of network communication and storage bandwidth in the computing system, therefore, the arithmetic throughput capacity and network communication and storage access pressure are considered, and the working efficiency of the whole system is improved.

The invention relates to a facilitate environmental control-oriented wireless sensor network data transmission method. The method comprises the following steps of processing the facilitate environmental data sequence acquired by a sensor by a sliding recursive method to obtain a sequence variance, detecting abnormal values in the facilitate environmental data sequence by the Letts' criterion; adaptively determining an error threshold allowable for sensor data of a facilitate environmental control system based on a parabolic equation; establishing a first-order linear regression model at a node side of the sensor and a server side and simultaneously fusing adaptive weighted algorithm in the server side to meet the condition of redundant deployment of sensor nodes to obtain fusion results of the sensor. According to the method, the data sending times of sensor nodes can be reduced to the maximum extent at a specified accuracy requirement, thereby effectively reducing the energy waste and data collisions.

The invention discloses a method and device for synchronizing meta information among digital nodes applied in a data storage system comprising multiple digital nodes. The method comprises: with respect to every data node, configuring a target data node which is in network connection with the data node and is used for synchronizing the meta information of the data node, wherein the all configured data nodes have network communication; detecting whether the metal information of the data node is updated, if so, taking the data node as a source data node, synchronizing the meta information of the source data node to the target data node, thus updating the meta information of the target data node. In adoption of the solution, the defects in the prior art that the meta information of all data nodes is needed to be synchronized, therefore the data transmission is frequent, the network cost is great and the expended time is long are overcome; the network cost is reduced; the transmission time is controlled; and the synchronizing efficiency is promoted.

The invention provides a monitoring video processing method and device, electronic equipment and a storage medium. The method comprises the following steps of: obtaining a monitoring video by the first articulated naturality web terminal; detecting a plurality of frames of video images included in the monitoring video; sending the detected monitoring video to a storage server; caching the detectedmonitoring video by the storage server to a cache region; when the first articulated naturality web terminal detects that the frame rate in the multiple frames of video images exceeds a threshold value, starting the first articulated naturality web terminal; and outputting an alarm message and sending a video recording instruction to a storage server, reading the first part of monitoring video from the cache region by the storage server according to the video recording instruction, receiving the second part of monitoring video by the storage server, and storing the first part of monitoring video and the second part of monitoring video. According to the invention, the camera is enabled to directly realize the application of the motion detection function in the articulated naturality web, the security of monitoring video data is improved, the working efficiency of the storage server is greatly improved, and the space of the storage server is saved.

The invention discloses a data aggregation method for increasing the capacity of a wireless sensor network. The method comprises the following steps of: homalographically clustering the wireless sensor network by using a circular grid; determining aggregated nodes of the clusters, a data aggregation mode of the network and a data transmission mode of the network; and creating a virtual node and thus acquiring a reachable throughput of the node by adopting a scheduling scheme with a load and correlation. By utilizing the method, an actual communication condition of the wireless sensor network is described by using a mode with the communication characteristic, closer to that of a sensor; and furthermore, the data transmission times in the network are reduced, therefore, the balance of the effectiveness and redundancy of information is realized, and the capacity of the wireless sensor network is increased.

The invention relates to a distributed filtering network system and a design method. The system comprises the following parts: a filter network unit comprising filters distributed at different physical sites; a sensor sampling unit comprising vehicle-mounted sensors and filtering sensors, wherein the vehicle-mounted sensors are embedded in an automobile suspension system and acquire state information of the automobile suspension system, and the filtering sensors are embedded into the filters and acquire exchange information between the filters and set adjacent filters; and an event trigger unit comprising event triggers corresponding to the filters, wherein each event trigger determines event occurrence time of a corresponding filter, at the event occurrence time, the filters receive the automobile suspension system state information acquired by the corresponding vehicle-mounted sensors and the exchange information of the set adjacent filters, and otherwise, the filters do not act. Compared to the prior art, the system provided by the invention can satisfy the demand of multiple terminals for simultaneously obtaining data, can also effectively prevent filter network collapse, and is low in energy consumption, high in efficiency and high in stability.

The invention discloses a method for proxying PIM at the forwarding layer: the source directly connects to the designated router and the forwarding layer stores the received multicast data in a temporary forwarding table item, obtains forwarding information corresponding to the multicast data from the routing protocol layer, and according to the forwarding information Whether there is a PIM registration message identifier to process the multicast data in the temporary forwarding table entry; the forwarding layer of the aggregation point router stores the PIM registration message received from the source directly connected to the designated router in the temporary forwarding table entry, from the routing protocol The layer obtains the forwarding information corresponding to the PIM registration message, and processes the PIM registration message in the temporary forwarding entry according to whether the received PIM registration message identifier is included in the forwarding information. In the embodiment of the present invention, aiming at the PIM registration message mechanism, the number of transfers of valid data at the PIM routing protocol layer and the forwarding layer is reduced, and flow control can be realized; especially, unnecessary inter-board communication in a distributed router is effectively avoided , improving system performance.