JP7056084B2 - Decoding device - Google Patents

Description

The present disclosure relates to a decoding device that decodes data encoded by using compressed sensing technology.

In recent years, compressed sensing technology has been attracting attention as a compression technology that can be applied to data having sparsity. For example, Patent Document 1 discloses a system for sharing an image captured by an in-vehicle camera compressed by using a compressed sensing technique by vehicle-to-vehicle communication.

As known decoding algorithms used in the technical field of compressed sensing, Matching Pursuit (MP), Orthogonal Matching Pursuit (OMP), Iterative Hard Thresholding (IHT), Weak MP, LS-OMP, Basis Pursuit (BP) ), Least Angle Regression (LARS), Iterative-Reweighted-Least-Squares (IRLS) Approximate Message Passing (AMP), and various other algorithms have been proposed. MP, OMP, IHT, Weak MP, and LS-OMP are decoding algorithms corresponding to the greedy algorithm. BP, IRLS, and LARS are a kind of convex relaxation method that approximates the cost function as a convex function.

Japanese Unexamined Patent Publication No. 2013-145954

Decoding processing in compressed sensing corresponds to an optimization problem in which an optimum solution cannot be obtained unless all combinations are tried, and NP-hardness is difficult. Therefore, as a realistic decoding algorithm, OMP, IHT, and the like have been proposed as described above. However, in these heuristic methods, it is assumed that the output solution contains an error. Therefore, as a decoding device used in the field of compressed sensing technology, a configuration for restoring original data with higher accuracy is required.

The present disclosure has been made based on this circumstance, and an object of the present invention is to provide a decoding device for more accurately decoding compression-encoded data using compressed sensing.

The decoding device according to the present disclosure for achieving the object is a decoding device that decodes compressed data which is data compressed by using compression sensing, and the compressed data has temporal continuity. Further, the uncompressed original data obtained by converting the state quantity data, which is the data indicating the state of a predetermined state quantity that changes dynamically at a certain time point, into one column vector having a predetermined number of elements, is further added. The data is compressed and encoded by a predetermined compression matrix, and each time the compressed data receiving unit (241) that sequentially receives the compressed data from another device and the compressed data receiving unit receives the compressed data, the received compression is performed. A decoding processing unit (242) that generates decoded data having the number of elements by executing a predetermined decoding algorithm on the data, and a restoration unit that restores state quantity data based on the decoding data generated by the decoding processing unit. (243), an estimation unit (246) that estimates the current state of the state quantity based on a plurality of state quantity data generated by the restoration unit within a certain period of time in the past, and one column vector having the number of elements. A filter generation unit (247) that generates a filter vector having a non-zero element at a position corresponding to the current state of the state quantity estimated by the distribution estimation unit is provided, and the decoding processing unit is provided by the filter generation unit. When a filter vector can be generated, an estimated solution corresponding to the original data is generated using the filter vector, compressed data, and a compressed matrix as a decoding algorithm, and the degree of deviation between the estimated solution and the original data is shown. It is configured to generate decoded data using an algorithm that repeatedly calculates the residual, which is an index, and corrects the estimated solution using a filter vector and a compression matrix until a predetermined end condition is satisfied. ..

When the filter generation unit can generate the filter vector, the decoding processing unit provided with the above configuration decodes the compressed data using the filter vector generated by the filter generation unit. The filter vector used here is a vector having a non-zero element at a position corresponding to the current state of the state quantity estimated by the distribution estimation unit. From another point of view, such a filter vector corresponds to a vector showing the result of estimating the position of the non-zero element in the uncompressed data of the received compressed data.

That is, the above configuration corresponds to a configuration in which the compressed data is decoded using the result of estimating the position of the non-zero element in the data before compression of the received compressed data. Further, according to the configuration in which decoding is performed using the result of estimating the position of the non-zero element in the data before compression, the information about the position where the non-zero element is arranged in the data before compression of the received compressed data can be obtained. It can be expected that the decoding accuracy will be higher than the configuration in which decoding is performed without using anything. Therefore, according to the above configuration, it is possible to more accurately decode the compressed coded data by using compressed sensing. The case where the filter generation unit can generate the filter vector corresponds to the case where at least one state quantity data generated by the restoration unit exists within the past fixed time. This is because the generation of the filter vector requires at least one of a plurality of state quantity data generated by the restoration unit within a certain period of time in the past.

The reference numerals in parentheses described in the claims indicate, as one embodiment, the correspondence with the specific means described in the embodiments described later, and limit the technical scope of the present disclosure. is not.

It is a conceptual diagram which shows the schematic structure of the communication system which concerns on this embodiment. It is a block diagram which shows the schematic structure of the transmission side system 1. It is a figure for demonstrating the peripheral object data. It is a figure which shows the matrix corresponding to the peripheral object data shown in FIG. It is a block diagram which shows the schematic structure of the receiving side system 2. It is a conceptual diagram for demonstrating the operation of the object tracking part 245. It is a conceptual diagram for demonstrating the operation of the distribution estimation part 246. It is a figure which shows an example of the occupancy probability map. It is a figure which shows an example of the occupancy probability map. It is a figure which shows an example of the occupancy probability map. It is a flowchart for demonstrating the restoration process which a decoding apparatus 24 carries out. It is a figure which summarized OMP. It is a figure which summarized the filter combined OMP as a proposed method. It is a figure which summarized the filter combined use IHT as a proposed method.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. FIG. 1 is a diagram showing an example of a schematic configuration of a communication system according to the present disclosure. As shown in FIG. 1, the communication system includes a transmitting side system 1 mounted on the vehicle Ma and a receiving side system 2 mounted on the vehicle Mb. For convenience, the vehicle Ma equipped with the transmission side system 1 is also referred to as a transmission vehicle Ma. Further, the vehicle Mb equipped with the receiving side system 2 is also referred to as a receiving vehicle Mb. When the transmitting vehicle Ma and the receiving vehicle Mb are not distinguished, it is described as a vehicle. Although only one transmitting side system 1 is shown in FIG. 1 for convenience, there may be a plurality of transmitting side systems 1. Along with this, there may be a plurality of transmitting vehicles Ma. Similarly, there may be a plurality of receiving system 2 and receiving vehicle Mb. The transmitting system 1 corresponds to another device for the receiving system 2.

The receiving vehicle Mb is a vehicle traveling on the road. The receiving vehicle Mb may be a two-wheeled vehicle, a three-wheeled vehicle, or the like, in addition to the four-wheeled vehicle. Motorcycles also include motorized bicycles. In the present embodiment, as an example, the receiving vehicle Mb is a four-wheeled vehicle. The transmitting vehicle Ma is also a vehicle traveling on the road like the receiving vehicle Mb, and the specific type thereof is not limited. As an example, the transmitting vehicle Ma is also a four-wheeled vehicle.

The transmitting side system 1 and the receiving side system 2 are configured to enable narrow-range communication using radio waves in a predetermined frequency band. Narrow-range communication here is direct wireless communication that does not go through a wide-area communication network. The frequency band used for narrow-range communication is, for example, the 760 MHz band. In addition, 2.4 GHz, 5.9 GHz band and the like can also be used. Any standard can be adopted as the standard that defines the frequency, modulation method, etc. for realizing narrow-range communication.

Here, as an example, based on the fact that both the transmitting side system 1 and the receiving side system 2 are mounted on the vehicle, the narrow range communication is a WAVE (Wireless Access in Vehicular Environment) standard disclosed in IEEE1609 and the like. It shall be carried out in accordance with. WAVE is a standard for direct wireless communication between vehicles (so-called vehicle-to-vehicle communication). The wireless communication between the transmitting side system 1 and the receiving side system 2 corresponds to vehicle-to-vehicle communication from another viewpoint.

Each of the plurality of vehicles sequentially broadcasts communication packets (hereinafter referred to as vehicle information packets) indicating vehicle information such as the current position, traveling speed, and traveling direction by vehicle-to-vehicle communication as narrow area communication, and others. The vehicle information packet transmitted from the vehicle is sequentially received. In addition to vehicle information, the vehicle information packet includes information such as the transmission time of the communication packet and source information. The source information is an identification number (so-called vehicle ID) assigned to the vehicle corresponding to the source.

Further, the transmitting side system 1 and the receiving side system 2 are the positions of objects such as vehicles, pedestrians, and falling objects existing around the transmitting vehicle Ma (in other words, the transmitting side system 1) by vehicle-to-vehicle communication as narrow area communication. Is configured to share peripheral object data expressed in grid map format. In other words, the peripheral object data is data showing the traffic condition and the traveling environment around the transmitting vehicle Ma. Traffic conditions change dynamically with temporal continuity. That is, the traffic condition around the transmitting side system 1 corresponds to a state quantity that dynamically changes while having temporal continuity. Further, the traveling environment of the transmitting side system 1 (positional relationship with a falling object or a structure, etc.) corresponds to a state quantity that dynamically changes while having continuity with the movement of the transmitting vehicle Ma. Therefore, the peripheral object data corresponds to the state quantity data. The periphery of the transmitting vehicle Ma here corresponds to a predetermined range determined based on the current position of the transmitting vehicle Ma.

<About the configuration of the transmitting system 1>
As shown in FIG. 2, the transmitting side system 1 includes a narrow area communication module 11, a peripheral monitoring sensor 12, a grid map generation unit 13, and a compression processing unit 14. The narrow area communication module 11 is a communication module for carrying out the above-mentioned narrow area communication. The narrow area communication module 11 is communicably connected to the compression processing unit 14. Further, the narrow area communication module 11 is also communicably connected to the grid map generation unit 13. The narrow-range communication module 11 modulates the data input from the compression processing unit 14 and the like, and wirelessly transmits the data via an antenna (not shown). Further, when the narrow area communication module 11 receives the vehicle information packet transmitted from another vehicle, the narrow area communication module 11 provides the data shown in the vehicle information packet (hereinafter, other vehicle data) to the grid map generation unit 13.

The peripheral monitoring sensor 12 is a sensor that detects an object existing in the vicinity of the transmitting vehicle Ma. The peripheral monitoring sensor 12 is, for example, a peripheral monitoring camera that captures a predetermined range around the transmitting vehicle Ma, a millimeter wave radar that transmits exploration waves to a predetermined range around the transmitting vehicle Ma, a sonar, and LIDAR (Light Detection and Ranging / Laser Imaging). Detection and Ranging) and other sensors. A peripheral surveillance camera is a device that outputs captured images that are sequentially captured as sensing information. A sensor that transmits exploration waves such as a sonar, a millimeter wave radar, and LIDAR is a sensor that outputs a scanning result based on a received signal y obtained when a reflected wave reflected by an obstacle is received as sensing information. The sensing information output by the peripheral monitoring sensor 12 is input to the grid map generation unit 13.

The grid map generation unit 13 sequentially identifies the positions of objects existing in the vicinity of the transmitting side system 1 based on the sensing information output by the peripheral monitoring sensor 12. As a more preferable embodiment, the grid map generation unit 13 of the present embodiment identifies the position of an object existing in the vicinity of the transmission side system 1 by using other vehicle data provided by the narrow area communication module 11. Then, as shown in FIG. 3B, data representing the positions of objects existing around the transmitting side system 1 in a grid map format (that is, peripheral object data) is generated. Objects here include pedestrians as well as vehicles. In other words, the peripheral object data can function as data indicating the space in which the vehicle can travel.

Peripheral object data is data that divides a predetermined range determined with reference to the transmitting system 1 into a plurality of cells and expresses whether or not an object exists in each cell by numerical values, colors, hatching patterns, and the like. be. FIG. 3A is a conceptual diagram showing a bird's-eye view of the traffic conditions around the transmitting side system 1 showing the degree of deviation between the estimated solution and the original data before compression, and FIG. 3B is a conceptual diagram. It is the peripheral object data corresponding to the traffic condition exemplified in (A) of. The hatched cell represents the cell in which the object exists (hereinafter referred to as the occupied cell). The cells that are not hatched represent cells in which no object exists (hereinafter referred to as blank cells). In FIG. 3B, whether or not an object exists is represented by the presence or absence of hatching.

The number of rows, the number of columns, and the total number of cells (hereinafter referred to as the total number of cells) N constituting the grid map as peripheral object data may be appropriately designed according to the size of the range represented by the grid map and the like. .. Here, as an example, it is assumed that the grid map is composed of cells having 6 rows and 10 columns. That is, the total number of cells N is 6 × 10 = 60. The size of the real space represented by the grid map may be appropriately designed according to the range regarded as the periphery of the transmitting vehicle Ma.

The size of the real space represented by the unit cell may also be appropriately designed. Here, as an example, one cell corresponds to a real space of 2 meters square. Since the grid map of the present embodiment has 6 rows in the vehicle width direction and 10 columns in the vehicle front-rear direction, the grid map is within a rectangular range of 12 m in the vehicle width direction and 20 m in the vehicle front-rear direction in the real space. It functions as data indicating the position of an existing object.

Here, for the sake of simplification of the explanation, the peripheral object data shall be data representing two states of whether or not an object exists in each cell, but the data is not limited to this. In another aspect, the peripheral object data may represent the probability that an object exists in each cell. In that case, a numerical value (for example, a decimal number such as 0.5) indicating the probability that the object exists may be inserted into each cell.

Further, the grid map as peripheral object data may be represented by a matrix as shown in FIG. 4, for example. The matrix as peripheral object data includes elements corresponding to each cell of the grid map. In the matrix as peripheral object data, the value of the element corresponding to the blank cell is set to 0, and the element corresponding to the occupied cell is set to a non-zero value (here, 1). That is, the element in which the non-zero value is set in the matrix shown in FIG. 4 corresponds to the occupied cell which is the hatched cell in the grid map shown in FIG. 3 (B).

The grid map as peripheral object data generated by the grid map generation unit 13 is provided to the compression processing unit 14. The grid map generation unit 13 specifies the positions of objects existing around the transmitting side system 1, and sequentially executes a process of generating a grid map as peripheral object data at predetermined generation intervals. The generation interval may be set to a value that does not significantly change the traffic conditions around the transmitting system 1. For example, the generation interval is preferably set to a value within 3 seconds, such as 100 milliseconds, 200 milliseconds, 500 milliseconds, 1 second, or 2 seconds. Here, as an example, it is assumed that the generation interval is set to 250 milliseconds. The generation interval may be appropriately designed according to the size of the real space represented by the grid map.

The compression processing unit 14 uses a predetermined compression matrix (so-called observation matrix) to generate compressed data which is data obtained by compression-encoding a grid map as peripheral object data. Specifically, it is as follows. The compression processing unit 14 converts the grid map as peripheral object data into one column vector x having the same number of elements (in other words, the number of dimensions) as the total number of cells N. The conversion from the grid map to the column vector may be realized by concatenating the column vectors constituting the grid map in the order of the column numbers. Of course, as another aspect, the column vector x may be generated by transposing an N-dimensional row vector formed by concatenating the row vectors constituting the grid map in the order of row numbers.

That is, the compression processing unit 14 converts the grid map as peripheral object data into an N-dimensional vector x. The N-dimensional vector x itself is also data showing the distribution of the object and corresponds to the original data in compressed sensing. The N-dimensional vector x formed by converting the grid map as the peripheral object data corresponds to the original data before compression. Then, the compression processing unit 14 generates an M-dimensional (M <N) vector y obtained by linearly transforming the N-dimensional vector x using a predetermined compression matrix. The compression matrix A is an M × N matrix, and the M-dimensional vector y corresponds to the compression data. The compressed data corresponds to a grid map as peripheral object data linearly transformed using a predetermined compression matrix A. The compression processing unit 14 outputs the generated compressed data to the narrow area communication module 11. As a result, the compressed data is broadcast at a cycle according to the generation interval.

The compressed data may be transmitted with information indicating the number of non-zero elements k included in the peripheral object data before compression added. According to such an aspect, the receiving side system 2 can specify the number k of non-zero elements of the received data. In view of such circumstances, as a more preferable embodiment in the present embodiment, it is assumed that information indicating the number of non-zero elements k included in the peripheral object data before compression is added to the compressed data as transmission data. Of course, the compressed data as transmission data does not have to include information indicating the number of non-zero elements k.

Each of the grid map generation unit 13 and the compression processing unit 14 is provided in the transmission side system 1 as a functional block realized by a CPU (not shown) executing predetermined software. The grid map generation unit 13 and the compression processing unit 14 may be realized as hardware. The embodiment realized as hardware also includes the embodiment realized by using one or more ICs.

<About the configuration of the receiving system 2>
Next, the configuration of the receiving side system 2 will be described with reference to FIG. As shown in FIG. 5, the receiving side system 2 includes a narrow area communication module 21, a map data storage unit 22, a locator 23, a decoding device 24, and a grid map storage unit 25. The narrow area communication module 21 is a communication module for carrying out the above-mentioned narrow area communication. The narrow area communication module 21 is communicably connected to the decoding device 24. Upon receiving the compressed data sequentially transmitted from the transmitting side system 1, the narrow area communication module 21 provides the compressed data to the decoding device 24. When the narrow area communication module 21 receives the vehicle information packet transmitted from another vehicle, the narrow area communication module 21 converts the data shown in the vehicle information packet (hereinafter referred to as other vehicle data) into the decoding device 24 or the receiving vehicle Mb. It is provided to a predetermined ECU (Electronic Control Unit) mounted on the vehicle.

The map data storage unit 22 is a non-volatile memory that stores map data indicating road connection relationships and the like. The map data has, for example, node data relating to a point (hereinafter referred to as a node) at which a plurality of roads intersect, merge, or branch, and link data relating to a road connecting the points (hereinafter referred to as a link). The link data includes node coordinates (latitude / longitude) between the start and end of the link, as well as the length and width of the link, the number of lanes, the road type, and the like. The map data may be configured to include a three-dimensional map consisting of a road shape and a point cloud of feature points of a structure. Further, the receiving side system 2 may be configured to acquire map data from an external server or the like.

The locator 23 is a device that calculates the current position of the receiving vehicle Mb. The locator 23 includes, for example, a GNSS (Global Navigation Satellite System) receiver and an inertial sensor. The GNSS receiver receives positioning signals from a plurality of artificial satellites. The inertial sensor includes, for example, a gyro sensor and an acceleration sensor. The locator 23 sequentially positions the current position of the receiving vehicle Mb by combining the positioning signal received by the GNSS receiver and the measurement result of the inertial sensor. The locator 23 may be configured to complementarily use the vehicle speed information, the mileage, and the like sequentially output from the vehicle speed sensor for positioning. The positioning result of the locator 23 is provided to the decoding device 24.

The decoding device 24 is configured to generate decoded data by executing a predetermined decoding algorithm on the compressed data received by the narrow area communication module 21. The decoded data is data obtained by decoding (in other words, restoring) an M-dimensional vector as compressed data into an N-dimensional vector. The decoded data indirectly shows the distribution of objects around the transmitting vehicle Ma, similar to the original data expressed as an N-dimensional vector x.

The decoding device 24 is realized by using a computer including a CPU, RAM, flash memory, and the like. The decoding device 24 includes a compressed data receiving unit 241, a decoding processing unit 242, a grid map restoration unit 243, a storage processing unit 244, an object tracking unit 245, a distribution estimation unit 246, and a filter generation unit 247 as more detailed functions. Each function of the decoding device 24 is realized by a CPU (not shown) executing predetermined software. Note that some or all of the functions of the decoding device 24 may be realized as hardware. The embodiment realized as hardware also includes the embodiment realized by using one or more ICs.

The compressed data receiving unit 241 is configured to acquire the compressed data received by the narrow area communication module 21. As described above, the compressed data here is an M-dimensional vector y obtained by compression-coding an N-dimensional vector showing the distribution of objects around the transmitting vehicle Ma with a compression matrix A.

The decoding processing unit 242 is configured to decode the compressed data by using a predetermined decoding algorithm. The details of the operation of the decoding processing unit 242 will be described later separately. The decoding processing unit 242 generates an N-dimensional vector z corresponding to an estimated solution of the original data from the M-dimensional vector as the received data by executing a predetermined decoding algorithm. This N-dimensional vector z corresponds to the decoded data of the received compressed data. The N-dimensional vector z as the decoded data generated by the decoding processing unit 242 is provided to the grid map restoration unit 243.

The grid map restoration unit 243 converts the N-dimensional vector z into a grid map. As a result, the grid map as the peripheral object data generated by the transmitting side system 1 is restored. The conversion from the N-dimensional column vector z to the grid map (in other words, restoration) may be performed by the reverse procedure of the conversion from the grid map to the N-dimensional column vector. The grid map restoration unit 243 corresponds to the restoration unit.

The storage processing unit 244 associates the grid map generated by the grid map restoration unit 243 with information indicating the source of the grid map (actually, the compressed data) (that is, source information) and reception time information to grid. It is configured to be stored in the map storage unit 25.

The object tracking unit 245 assigns a unique object number to each object shown in the received grid map, and tracks the same object by using a well-known object tracking method. The received grid map refers to a grid map reproduced from the compressed data received by the compressed data receiving unit 241 in cooperation (in other words, cooperation) between the grid map restoration unit 243 and the decoding processing unit 242. ..

FIG. 6 is a diagram conceptually showing the operation of the object tracking unit 245. The ID in the figure represents an object number. In this way, each object is identified using the object number, and the previously received grid map and the newly received grid map are associated with each other. Then, the relative moving direction and moving speed of the object are specified from the degree of change in the position of the same object between a plurality of consecutive frames. The moving direction and moving speed of each object may be held in association with the object number.

The distribution estimation unit 246 estimates the current position of an object existing around the transmission vehicle Ma based on a plurality of grid maps from the same transmission vehicle Ma generated by the grid map restoration unit 243 within a certain period of time in the past. .. The distribution estimation unit 246 corresponds to the estimation unit.

For example, as shown in FIG. 7, when the object to which the object number 001 is assigned moves by one cell on the right side of the page each time the distribution estimation unit 246 receives the grid map, the current position is received last time. It is estimated that the position is moved by one cell to the right of the position shown on the grid map. If an object moves two cells to the right of the page each time it receives a grid map, the current position is the position moved two cells to the right of the position shown in the previously received grid map. Presumed to be. In the figure, t represents the reception time of the previous grid map, and t-1 represents the reception time of the grid map two times before. In FIG. 7, t + 1 represents the current time.

The distribution estimation unit 246 estimates the current position of an object existing within a predetermined range around the transmitting vehicle Ma by performing the above-mentioned estimation process of the current position for each object. The current position of each object can be estimated based on the moving direction and moving speed specified by the object tracking unit 245.

Then, the distribution estimation unit 246 generates a grid map (hereinafter, occupancy probability map) showing the probability that the object exists in a predetermined range corresponding to the periphery of the transmitting vehicle Ma. The occupancy probability map has the same number of rows and columns as the grid map generated by the transmitting vehicle Ma. That is, the configuration of the occupancy probability map itself is the same as the grid map generated by the transmitting vehicle Ma. Therefore, the number of cells included in the occupancy probability map is also N = 60.

In each cell constituting the occupancy probability map, a numerical value indicating the probability that an object exists in the cell (that is, the occupancy probability) is inserted. A cell whose occupancy probability is set to 0 means that it is estimated that there is no object in the cell. In the occupancy probability map, the cell corresponding to the current position of each object estimated based on the received grid map is a non-zero element.

The occupancy probability of a cell corresponds to the existence probability, which is the probability that an object exists in the cell. The distribution estimation unit 246 estimates the current position for each of the objects that can be tracked, and calculates the existence probability of each cell for each object. The existence probability of an object in a cell is the probability that the object exists in the cell. The existence probability of each cell for a certain object is determined according to the current position of the object and the like. It is preferable to set the existence probability of each cell of the object to a higher value as the cell is located in the traveling direction of the object to be detected (in other words, tracked).

For example, since the object with the object number 001 shown in FIG. 7 is moving to the right of the paper surface, the cell adjacent to the left-right direction is set to a higher value than the cell adjacent to the cell vertically to the estimated current position. .. FIG. 8 shows the occupancy probability map corresponding to the estimation result shown in FIG. 7 in a matrix format for convenience. The broken line in FIG. 8 represents the current position of the object with the object number 001 estimated by the distribution estimation unit 246.

Further, considering that the object with the object number 001 is moving to the right side of the paper, the occupancy probability map corresponding to the estimation result shown in FIG. 7 is a cell adjacent to the left side of the estimated current position as shown in FIG. It may be set so that the cell adjacent to the right side has a higher value than the value. In this embodiment, the existence probability of each cell of the object is calculated in consideration of the moving direction of the object.

Further, it is preferable that the existence probability of each cell of the object is set by distributing it over a wide range along the moving direction as the moving speed of the object increases. For example, the object with the object number 001 as an object is assumed to be moved to the right by one cell in FIG. 7, but if it is moved to the right by two cells, it is estimated as shown in FIG. A non-zero value is set as the occupancy probability from the current position to the two cells on the right side.

The distribution estimation unit 246 generates an occupancy probability map showing the occupancy probability for each cell by adding the existence probabilities of each cell for each object calculated by the above rule. For cells in which the existence probabilities of a plurality of objects overlap each other, the sum of the existence probabilities of the respective objects may be set as the occupancy probability of the cell. That is, the distribution estimation unit 246 generates an occupancy probability map by calculating the existence probability of each cell according to the current position of each object. The occupancy probability map corresponds to data indicating the probability that each cell is an occupant cell in the grid map as the original data of the received signal y.

The filter generation unit 247 generates a filter vector F obtained by converting the occupancy probability map generated by the distribution estimation unit 246 into an N-dimensional column vector. Since the occupancy probability map itself is data representing the existence probability of an object for each point (specifically, for each cell) around the transmitting vehicle Ma, this filter vector F also corresponds to a point corresponding to the peripheral region of the transmitting vehicle Ma. The existence probability of each object is shown. The filter vector F includes a non-zero element at a position corresponding to the distribution of objects around the transmitting vehicle Ma.

The vector x as the original data corresponding to the received signal y is data indicating the distribution of objects existing around the transmitting vehicle Ma, and the filter vector F is for each point corresponding to the peripheral region of the transmitting vehicle Ma. It is the data which shows the existence probability of the object of. Then, the points indicated by the elements of the filter vector F and the vector x in the real space correspond to each other.

That is, according to another viewpoint, the filter vector F corresponds to a vector showing the result of estimating the position of the non-zero element in the vector x as the original signal of the received signal y. The numerical value for each element included in the filter vector F corresponds to the probability that the element has a non-zero value set in the vector x as the original signal. The filter vector F generated by the filter generation unit 247 is used by the decoding processing unit 242.

The grid map storage unit 25 stores the grid map as peripheral object data received from the transmission vehicle Ma as another vehicle. The grid map storage unit 25 is realized by using a rewritable storage medium such as a RAM or a flash memory. The grid map is stored in association with information indicating the source of the compressed data (that is, source information) and reception time information. The grid map is saved for at least a predetermined time from the reception time, and then deleted at any time. The storage period is, for example, 30 seconds or 1 minute.

<Decryption process>
Next, the restoration process performed by the decoding device 24 will be described with reference to the flowchart shown in FIG. The restoration process is a process for restoring the grid map from the compressed data acquired by the compressed data receiving unit 241. As shown in FIG. 6, the restoration process includes steps S101 to S107. The restoration process may be started, for example, when the compressed data receiving unit 241 receives the compressed data from the narrow area communication module 21 as a trigger.

First, in step S101, there are two grid maps received from the same transmission vehicle Ma as the source of the compressed data received this time, and the reception time is within a predetermined extraction condition time from the present. It is determined whether or not the data is stored in the grid map storage unit 25. The extraction condition time may be appropriately designed according to the grid map generation interval in the transmitting vehicle Ma, and here, it is assumed that the extraction condition time is set to a value four times the generation interval, that is, 2 seconds. In this step S101, whether or not the distribution estimation unit 246 can estimate the current position of the object existing around the transmitting vehicle Ma based on the received grid map, in other words, the filter vector F can be generated. It corresponds to the process of determining whether or not.

In step S101, the grid map received from the same transmitting vehicle Ma as the source of the compressed data (that is, the received signal y) received this time, and the reception time is within the extraction condition time from the present. If two or more are present, step S103 is executed. On the other hand, there are 0 or 1 grid maps received from the same transmission vehicle Ma as the source of the compressed data received this time, and the reception time is within the extraction condition time from the present. In that case, step S102 is executed.

In this embodiment, in order to generate an occupancy probability map (and thus a filter vector F) using the moving direction and moving speed of the object, it is required that there are two or more grid maps received within the extraction condition time from the present time. , It is a condition for generating an occupancy probability map, but it is not limited to this. When generating the occupancy probability map without considering the movement speed and the movement direction, at least one grid map from the same transmission vehicle Ma as the source of the compressed data received this time is within the extraction condition time from the present time. It suffices if it can be received. That is, in step S101, one or more grid maps are grid maps received from the same transmission vehicle Ma as the source of the compressed data received this time, and the reception time is within the extraction condition time from the present. It may be configured to determine whether or not it is stored in the map storage unit 25.

In step S102, the decoding processing unit 242 decodes the compressed data as the received signal y by using a predetermined decoding algorithm known in the technical field of compressed sensing. Known decoding algorithms used in the field of compressed sensing include Matching Pursuit (MP), Iterative Hard Thresholding (IHT), Orthogonal MP (so-called OMP), Weak MP, LS-OMP, Basis Pursuit (BP), etc. Iterated-Reweighted-Least-Squares (IRLS), Least Angle Regression (LARS), Approximate Message Passing (AMP), etc. Note that MP, IHT, OMP, Weak MP, and LS-OMP are decoding algorithms corresponding to the greedy algorithm. BP, IRLS, and LARS are a kind of convex relaxation method that approximates the cost function as a convex function.

Here, as an example, the decoding processing unit 242 decodes the received signal y using OMP. Roughly speaking, the OMP has k pieces having a strong correlation with the residual r determined based on the received signal y among the N column vectors _ai (i = 1, 2, ..., N) included in the compression matrix A. This is a method of restoring the vector x (that is, the original data) as the original data by using the column vector (k is a natural number).

FIG. 12 summarizes the OMP algorithm, and the specific procedure is as shown in FIG. As shown in FIG. 12, the OMP includes processes 1 to 4. The process 1 is a process of updating the support set S, which is a set of element numbers for which a non-zero value is set in the original data. In process 1, the column vector ai having the highest correlation with the residual r (the absolute value of the inner product is large) is specified in the compression matrix A, and the subscript _i of the column vector _ai is added to the support set S. The apparent variable added to the support set S in process 1 is j, but the variable j is set to the subscript i that satisfies the above conditions. Therefore, the process 1 is a process of adding the subscript _i of the column vector ai that substantially satisfies the above condition to the support set S. Note that i and j are variables set to integers from 1 to N.

Process 2 is a process of calculating the best solution in the support set S. The process 3 is a process of calculating the residual provided by the estimated solution calculated in the process 2, and the process 4 is a process of determining whether or not the result of the process 3 satisfies the end condition. The termination conditions may be appropriately designed. Here, it is assumed that the repetition process is terminated when the residual r is less than the approximate threshold value ε or when the repetition number p exceeds the repetition upper limit value q. Specific values of the approximate threshold value ε and the repeat upper limit value q may be appropriately designed.

The residual r in the OMP is the result of subtracting a signal (specifically, an M-dimensional column vector) obtained by compressing and converting a provisional estimated solution with a compression matrix A from the received signal y. The residual r is a parameter that functions as an index indicating the degree of deviation between the estimated solution and the original data. In the compression matrix A, the column vector having a strong correlation with the residual r determined based on the received signal y is a column vector having a large absolute value of the inner product with the residual r. That is, OMP is a method of adopting as a support in order from the column having the largest absolute value of the inner product with the residual r. As the number of non-zero elements k, the one added to the compressed data may be used. Further, the number of non-zero elements k may be a preset value. When the arithmetic processing in step S102 is completed, step S106 is executed.

In this embodiment, as an example, when the current position of an object existing around the transmitting vehicle Ma cannot be estimated based on the received grid map, the compressed data is decoded using OMP. Not limited to. As a known decoding algorithm (hereinafter referred to as an alternative algorithm) used when the current position of an object existing around the transmitting vehicle Ma cannot be estimated based on the received grid map, an algorithm other than OMP can also be adopted. For example, the alternative algorithm may be IHT. Further, the alternative algorithm may be BP or the like. Various known decoding algorithms can be adopted.

Generally, the IHT calculates a residual r for an estimated solution that is tentatively decoded data, and repeats a process of correcting the estimated solution by the residual r to obtain a final solution (that is, decoding). It is an algorithm to obtain data). In IHT, in the phase of correction of the estimated solution, the absolute value of the elements of the corrected estimated solution is large in order to keep the estimated solution k-sparse, assuming that the number of non-zero elements k of the original data is known. Set to 0 except for k. The residual r is a vector obtained by subtracting a signal obtained by linearly transforming an estimated solution using a compression matrix from a received signal y as compressed data.

In step S103, the distribution estimation unit 246 creates an occupancy probability map showing the existence probability of an object at each corresponding point around the transmitting vehicle Ma based on a plurality of grid maps received from the transmitting vehicle Ma within the past fixed time. Generate. When the process in step S103 is completed, step S104 is executed.

In step S104, the filter generation unit 247 generates a filter vector F indicating the existence probability of an object at each corresponding point around the transmitting vehicle Ma based on the occupancy probability map generated by the distribution estimation unit 246. When the process in step S104 is completed, step S105 is executed.

In step S105, the decoding processing unit 242 decodes the compressed data as the received signal y by using the algorithm shown in FIG. 13 as the proposed method. The algorithm as the proposed method in the present embodiment is a novel algorithm based on OMP, and is referred to as OMP with a filter for convenience. FIG. 13 summarizes the OMP with a filter, and the specific procedure is as shown in FIG. As shown in FIG. 13, the filter combined OMP includes processes 1 to 4. The process 1 is a process of updating the support set S, in which the column vector _ai having the maximum value obtained by multiplying the inner product with the residual r by the value fi of the i-th element of the filter vector F in the compression matrix A is performed. Add the subscript i of to the support set S. The processes 2 to 4 are the same as the OMP processes 2 to 4 shown in FIG.

That is, the difference between the OMP with a filter and the conventional OMP is that the weighting by the filter vector F is introduced in the selection (update) of the support set in the process 1 as follows. In OMP, the subscript _i of the column vector ai having the highest correlation with the residual r (the absolute value of the inner product is large) in the compression matrix A is added to the support set S. On the other hand, in the proposed method, in the compression matrix A, the subscript _i of the column vector ai that maximizes the value obtained by multiplying the inner product with the residual r by the value fi of the i-th element of the filter vector F is used as the support set S. to add.

According to such a configuration, it is possible to reduce the possibility that the subscript i of the column vector having a weak correlation with the vector x as the original signal is selected as the support in the compression matrix A. That is, there is a high possibility that the column vector component having a weak correlation with the vector x as the original data can be excluded from the process of optimizing the estimated solution. As a result, the estimation error can be suppressed.

In step S106, the grid map restoration unit 243 converts the N-dimensional vector z as the decoded data obtained in the above processing into a grid map, and in step S107, the storage processing unit 244 converts the grid map generated in step S106 into a grid map. The data is stored in the grid map storage unit 25 in association with the source information and the reception time information, and this flow is terminated.

The grid map as peripheral object data handled in this embodiment has a high spatial correlation on the time axis. The present disclosure focuses on the spatiotemporal correlation of the data to be transmitted and received, and the decoding accuracy is improved by using the filter vector F determined by utilizing the spatiotemporal correlation of the data to be transmitted and received. Increase. Further, in the present embodiment, by incorporating the moving feature amount of the object (for example, the moving speed and the moving direction) into the filter vector F, more accurate decoding becomes possible.

By the way, in the present embodiment, as an example, the filter combined OMP is used as an algorithm for correcting the estimated solution by using the filter vector F and the compression matrix, but the present invention is not limited to this. The filter vector F can be applied to various decoding algorithms and contributes to the improvement of the decoding accuracy and the convergence speed to the optimum solution. For example, the method of updating the support set S using the weighting by the filter vector F is also applicable to MP, which is the algorithm underlying OMP. It can also be applied to other MP-based algorithms (eg Weak MP, LS-OMP, etc.).

Furthermore, updating the estimated solution using the filter vector F is also applicable to IHT. FIG. 14 shows, as the second proposed method, an algorithm (hereinafter, filter-combined IHT) in which weighting processing by the filter vector F is applied to the update of the estimated solution in the IHT. The filter combined IHT includes processes 1 to 5 as shown in FIG. Process 1 is a process of calculating the basic update vector v. The basic update vector is a vector in which the residual r is linearly mapped by the transposed matrix of the compression matrix A. In the process 2, the basic update vector v generated in the process 1 is weighted by the filter vector F, and the weighted update vector c is generated. In process 3, the estimated solution is updated using the weighted update vector generated in process 2. Process 4 is a process of calculating the residual r. Process 5 is a process of determining whether or not the end condition is satisfied.

The differences between the IHT with a filter as the proposed method and the normal IHT are as follows. In a normal IHT, among the elements included in the vector obtained by adding the basic update vector v defined in FIG. 14 to the estimated solution, the process of setting the elements other than k having a large absolute value to 0 is repeated. So, I will update the estimated solution. On the other hand, in the second proposed method, as defined as process 2, a vector obtained by multiplying each element of the basic update vector v and each element of the filter vector F is added to the estimated solution, and the vector is added to the estimated solution. Among the elements included in the vector after addition, the estimated solution is updated by repeating the process of setting the elements other than k having a large absolute value to 0. Even in such an aspect, the same effect as the above-mentioned OMP with a filter can be obtained. An algorithm that repeatedly performs a process of updating an estimated solution using a filter vector F and a compression matrix A, such as an OMP with a filter and an IHT with a filter, corresponds to an algorithm with a filter.

Although the embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments, and various modifications described below are also included in the technical scope of the present disclosure, and other than the following. Can be changed and implemented within the range that does not deviate from the gist. For example, the following various modifications can be appropriately combined and carried out within a range that does not cause a technical contradiction.

The members having the same functions as those described in the above-described embodiment are designated by the same reference numerals, and the description thereof will be omitted. Further, when only a part of the configuration is referred to, the configuration of the embodiment described above can be applied to the other parts.

[Modification 1]
In the proposed method described above, when the number of repetitions p exceeds a predetermined filter release threshold value, decoding with a normal OMP or a normal IHT is performed using the estimated solution at that time (in other words, as an initial solution). It may be controlled to continue. That is, in step S105, the estimated solution is calculated by the proposed method (for example, OMP with a filter or IHT with a filter) until the number of repetitions p exceeds the filter cancellation threshold value, while the number of repetitions p exceeds the filter cancellation threshold value. In, the estimated solution at that time is taken over and the calculation of the estimated solution by ordinary OMP or the like is continued. Such a configuration corresponds to a configuration in which the proposed method and the conventional method are combined for decoding.

The actions and effects of Modification 1 are as follows. The proposed method uses the filter vector F to promote convergence to the optimum solution, but can also act in the direction of excluding the existence of a new object that has not been detected so far. According to the configuration of this modification for such a problem, it is possible to flexibly deal with the appearance of a new object that has not been detected so far. That is, it is possible to achieve both the effects described in the above-described embodiment and the flexibility of dealing with newly detected objects.

The new object that has not been detected so far is an object that is not represented by the saved grid map, for example, an object that has entered the real space represented by the grid map, or is being tracked (in other words, detected). It is an object separated from the object (already done). The object separated from the object being tracked is an object falling from the vehicle as an object being tracked, a person who got off the vehicle, or the like.

The filter release threshold applied in the above configuration may be appropriately designed in a range smaller than the repeat upper limit value q. For example, the filter release threshold value may be set to a value such as one-half, one-third, one-fourth of the repetition upper limit value q. The filter release threshold value may be dynamically set according to the number of non-zero elements k of the received grid map in order to suppress an excessive filter effect. For example, the smaller the number of non-zero elements k, the smaller the filter release threshold value is set.

[Other variants]
-The device that distributes peripheral object data may be a device installed on / along the road (so-called roadside machine). In that case, the peripheral mobile object data (strictly speaking, the compressed data thereof) generated by the roadside machine may be provided to the receiving side system 2 by road-to-vehicle communication.
-In the above decoding process, the embodiment in which a plurality of decoding algorithms are executed in order is disclosed, but the present invention is not limited to this. A plurality of decoding algorithms may be executed in parallel. Further, a plurality of decoding algorithms may be executed in parallel, and an error may be detected by a majority vote.
-The receiving side system 2 and the transmitting side system 1 may be mounted on one vehicle. From another point of view, such a configuration corresponds to a configuration in which the receiving side system 2 including the decoding device 24 also has the function of the transmitting side system 1.
In the above, the aspect of compressing and encoding the data indicating the position of the moving body existing around the transmitting side system 1 and transmitting the data is disclosed, but the data to be compressed and encoded is not limited to this. The above configuration is applied to data at a given point in time for a given state quantity that changes dynamically while having temporal continuity, such as a grid map showing the distribution of temperature and rainfall. be able to.
-The receiving side system 2 does not have to be mounted on the vehicle, and may be built on the roadside machine. Further, the receiving side system 2 may be a mobile terminal such as a smartphone or a wearable device.
-Even if the transmission and reception of compressed data is performed in accordance with known short-range wireless communication standards such as Wi-Fi (registered trademark) (IEEE802.11a / b / g / n / p) and Bluetooth (registered trademark). good. Further, transmission / reception of compressed data may be performed via a wide area communication network.

1 Transmitter system, 2 Receiver system, 11 Narrow area communication module, 12 Peripheral monitoring sensor, 13 Original data generation unit, 14 Compression processing unit, 21 Narrow area communication module, 22 Map data storage unit, 23 Locator, 24 Decoding device , 25 Grid map storage unit, 241 Compressed data reception unit, 242 Decoding processing unit, 243 Grid map restoration unit (restoration unit), 244 Storage processing unit, 245 Object tracking unit, 246 Distribution estimation unit (estimation unit), 247 Filter generation Department

Claims (6)

A decoding device that decodes compressed data, which is compressed data using compressed sensing.
The compressed data is a single column vector having a predetermined number of elements of state quantity data, which is data indicating a state at a certain time point for a predetermined state quantity that dynamically changes while having temporal continuity. The original data before compression, which is converted to, is further compressed and encoded by a predetermined compression matrix.
A compressed data receiving unit (241) that sequentially receives the compressed data from another device, and
Decoding processing unit (242) that generates decoding data having the number of elements by executing a predetermined decoding algorithm for the received compressed data each time the compressed data receiving unit receives the compressed data. When,
A restoration unit (243) that restores the state quantity data based on the decoding data generated by the decoding processing unit, and
An estimation unit (246) that estimates the current state of the state quantity based on a plurality of state quantity data generated by the restoration unit within a certain period of time in the past.
A filter generation unit (247) that generates a filter vector having a non-zero element at a position corresponding to the current state of the state quantity estimated by the estimation unit, which is one column vector having the number of elements. Equipped with
When the filter generation unit can generate the filter vector, the decoding processing unit corresponds to the pre-compression original data using the filter vector, the compression data, and the compression matrix as the decoding algorithm. The estimated solution is generated, the residual, which is an index indicating the degree of deviation between the estimated solution and the original data before compression, is calculated, and the filter vector and the compression matrix are used until a predetermined end condition is satisfied. A decoding device configured to generate the decoded data using an algorithm that repeatedly performs a process of correcting the estimated solution. The decoding device according to claim 1.
The state quantity data is peripheral object data representing the presence or absence of an object or the existence probability of an object at each point within a predetermined range determined according to the current position of the other device in a grid map format.
The decoding processing unit converts the decoded data into the peripheral object data, and then converts the decoded data into the peripheral object data.
The estimation unit estimates the distribution of the object within the predetermined range based on the plurality of peripheral object data generated by the restoration unit within a certain period of time in the past.
The filter generation unit is a decoding device configured to generate a vector having a non-zero element at a position corresponding to the distribution of the object estimated by the estimation unit as the filter vector. The decoding device according to claim 2.
The estimation unit identifies the movement direction of each object shown in the peripheral object data based on the plurality of peripheral object data generated by the restoration unit within a certain period of time in the past, and determines the movement direction of each object in the specified movement direction. Based on this, a decoding device configured to estimate the current position of each object within the predetermined range. The decoding device according to claim 3.
The estimation unit specifies the moving speed of each object shown in the peripheral object data based on a plurality of peripheral object data generated by the restoration unit within a certain period of time in the past, and the moving direction of each object. And a decoding device configured to estimate the current position of each object within the predetermined range based on the moving speed. The decoding device according to any one of claims 1 to 4.
The decoding processing unit is
If the filter generator is not capable of generating the filter vector, it may use any of the Orthogonal Matching Pursuit (OMP), Matching Pursuit (MP), and Iterative Hard Thresholding (IHT) used in the compressed sensing. While generating the decoded data,
When the filter generation unit can generate the filter vector, the integer j satisfying the following equation in the compression matrix is the number of the element in which the non-zero value is set in the original data before compression. By adding to the support set, which is a set, the estimated solution is generated, and the residual, which is an index indicating the degree of deviation between the estimated solution and the original data before compression, is calculated, and a predetermined end condition is satisfied. A decoding device configured to generate the decoding data by using a filter combined algorithm that repeatedly performs a process of correcting the estimated solution using the filter vector and the compression matrix.

The decoding device according to claim 5.
When the filter generation unit can generate the filter vector and the number of repetitions of the processing exceeds a predetermined filter release threshold value in the filter combined algorithm, the decoding processing unit includes the case. A decoding device configured to generate the decoding data using any of the OMP, the MP, and the IHT with the estimated solution obtained at a time point as an initial solution.

Images