Self-adaptive three-dimensional array indoor positioning method and system

Abstract

The invention discloses a self-adaptive three-dimensional array indoor positioning method and system, and the method comprises the steps: dividing a to-be-positioned indoor region into a plurality of square sub-regions, and collecting the environment data of each square sub-region; the environment data comprises signal intensity RSSI data and light intensity data; preprocessing the collected environment data, and constructing a training set, a verification set and a test set; training the built LoRES network by using the training set, detecting the positioning precision of the trained network by using the verification set after each training, and testing the trained network with the positioning precision reaching the standard by using the test set to finally obtain an indoor positioning network; and the indoor positioning network determines the position information of the square sub-region where the indoor positioning network is located by acquiring the environment data of the square sub-region where the indoor positioning network is located.

Description

Technical Field

[0001] The present invention belongs to the field of complex indoor scene positioning and wireless communication technology, and specifically relates to an adaptive stereo array indoor positioning method and system. Background Art

[0002] In the 6G communication environment, the demand for high-precision indoor positioning is becoming more and more significant. In AR / VR applications, users need accurate location and environmental information to achieve seamless interaction; in industrial environments, accurate positioning of equipment can improve production efficiency and safety; smart home systems rely on accurate user positioning to provide personalized services. At present, satellite-based positioning services can complete high-precision positioning of outdoor environments, but due to the lack of satellite positioning system signals, the accuracy of indoor positioning still needs to be further improved. Fingerprint positioning has recently attracted attention due to its good performance, such as radio frequency identification, ZigBee, Bluetooth, WiFi, etc. Due to the relatively low cost of WiFi fingerprint infrastructure and user equipment, it has become a hot topic in indoor positioning research, but the positioning accuracy of WiFi fingerprint positioning is still needed to be improved due to interference from complex indoor environments. Summary of the invention

[0003] Purpose of the invention: In order to solve the problem of low positioning accuracy of existing indoor positioning methods caused by interference from complex indoor environments, the present invention proposes an adaptive stereo array indoor positioning method and system, which fully considers the changes in the indoor environment and the adaptability of the positioning environment, and effectively solves the positioning problems caused by high-frequency personnel flow and multiple regional similarities.

[0004] Technical solution: An adaptive stereo array indoor positioning method, comprising the following steps: Divide the indoor area to be located into several square sub-areas, and collect environmental data of each square sub-area; the environmental data includes signal strength RSSI data and light intensity data; Preprocess the collected environmental data and construct training sets, validation sets, and test sets; The training set is used to train the constructed Lo_RES network. After each training, the validation set is used to detect the positioning accuracy of the Lo_RES network after training. The test set is used to test the trained Lo_RES network with the positioning accuracy reaching the standard, and finally the indoor positioning network is obtained. The indoor positioning network obtains environmental data to determine the square sub-area where it is located; The Lo_RES network includes: a first convolutional layer, a first residual network block, a second residual network block, a third residual network block and a fully connected layer; the output of the first convolutional layer is input to the first residual network block, the input of the first residual network block and its output are added on the channel and input to the second residual network block, the input of the second residual network block and its output are added on the channel and input to the third residual network block, the output of the third residual network block is connected to the fully connected layer after a global average pooling, the corresponding probability value of the square sub-region where the fully connected layer outputs is located, and the center coordinates of the square sub-region with the largest probability value are taken as the final inferred position information; Each residual network block consists of a second convolutional layer, a first batch normalization layer, an activation layer, a third convolutional layer, and a second batch normalization layer connected in sequence.

[0005] Furthermore, the indoor area to be located is divided into a number of square sub-areas, and the environmental data of each square sub-area is collected; the environmental data includes signal strength RSSI data and light intensity data, which can be replaced by the following steps: The indoor area to be located is divided into several square sub-areas, and the 24 hours of the day are divided into multiple collection time periods; in each collection time period, the environmental data corresponding to each square sub-area is collected.

[0006] Furthermore, in each collection period, collecting the environmental data corresponding to each square sub-area specifically includes: According to the environmental characteristics of each collection period, the proportion of signal strength RSSI data and light intensity data in the environmental data is adjusted to obtain the environmental data corresponding to each square sub-area.

[0007] Furthermore, the preprocessing of the collected environmental data includes: Combining the signal strength RSSI data into an image data, and performing normalization processing on the image data; The light intensity data is combined into another image data, and the image data is normalized.

[0008] Furthermore, in the Lo_RES network, the input data of the first convolutional layer is expressed as: ,in, Represents the input image data, 2 represents the number of channels of the input image data, and 1 represents the image format is grayscale.

[0009] Furthermore, the first convolutional layer is a convolutional layer with a convolution kernel size of 2×2, a number of convolution kernels of 6, a stride of 1, and a padding of 0.

[0010] Furthermore, the second convolution layer and the third convolution layer are both convolution layers with a convolution kernel size of 1×1, a number of convolution kernels of 24, a stride of 0, and a padding of 0.

[0011] The present invention also discloses an adaptive stereo array indoor positioning system, comprising: The three-dimensional array is used to obtain environmental data, input the environmental data into the indoor positioning network, and obtain its own location information; The indoor positioning network is trained according to the following steps: Divide the indoor area to be located into several square sub-areas, and collect environmental data of each square sub-area; the environmental data includes signal strength RSSI data and light intensity data; Preprocess the collected environmental data and construct training sets, validation sets, and test sets; The training set is used to train the constructed Lo_RES network. After each training, the validation set is used to detect the positioning accuracy of the Lo_RES network after training. The test set is used to test the trained Lo_RES network with the positioning accuracy reaching the standard, and finally the indoor positioning network is obtained. The Lo_RES network includes: a first convolutional layer, a first residual network block, a second residual network block, a third residual network block and a fully connected layer; the output of the first convolutional layer is input to the first residual network block, the input of the first residual network block and its output are added on the channel and input to the second residual network block, the input of the second residual network block and its output are added on the channel and input to the third residual network block, the output of the third residual network block is connected to the fully connected layer after a global average pooling, the corresponding probability value of the square sub-region where the fully connected layer outputs is located, and the center coordinates of the square sub-region with the largest probability value are taken as the final inferred position information; Each residual network block consists of a second convolutional layer, a first batch normalization layer, an activation layer, a third convolutional layer, and a second batch normalization layer connected in sequence.

[0012] Furthermore, the three-dimensional array includes: WiFi signal receiver module, used to collect signal strength RSSI data in environmental data; Light intensity tester module, used to collect light intensity data from environmental data; The calculation module stores the indoor positioning network and obtains its own location information by using the signal strength RSSI data collected by the WiFi signal receiver module and the light intensity data collected by the light intensity tester module.

[0013] Furthermore, the indoor area to be located is divided into a number of square sub-areas, and the environmental data of each square sub-area is collected; the environmental data includes signal strength RSSI data and light intensity data, which can be replaced by the following steps: Divide the indoor area to be located into several square sub-areas, and divide the 24 hours of the day into multiple collection periods; According to the environmental characteristics of each collection period, the proportion of signal strength RSSI data and light intensity data in the environmental data is adjusted to obtain the environmental data corresponding to each square sub-area; The preprocessing of the collected environmental data specifically includes: combining the signal strength RSSI data into an image data, and normalizing the image data; combining the light intensity data into another image data, and normalizing the image data; In the Lo_RES network, the input data of the first convolutional layer is represented as: ,in, Represents the input image data, 2 represents the number of channels of the input image data, and 1 represents the image format is grayscale.

[0014] Beneficial effects: Compared with the prior art, the present invention has the following advantages: (1) Different from the traditional single information positioning system, the present invention adopts adaptive and highly robust dual signal source positioning, fully considering the changes in the indoor environment and the adaptability of the positioning environment, effectively solving the positioning problems caused by high-frequency personnel flow and multiple regional similarities, and is suitable for positioning in GNSS-denied scenarios; (2) The present invention uses a machine learning algorithm to continuously optimize the network by adjusting network parameters, extracts the characteristics of complex indoor environment data in different time periods, and uses light intensity and signal strength RSSI data for joint collaborative positioning, thereby improving the adaptability of indoor positioning technology and the self-learning ability of neural networks, thereby improving the accuracy of complex indoor positioning; (3) Since the flow of people indoors has a great impact on the signal strength RSSI data, and the occlusion and reflection of indoor objects can cause the signal strength RSSI data in different areas to be similar in size, the method of the present invention, supplemented by light intensity information, can help to distinguish the specific location indoors. The light intensity data and signal strength RSSI data are two independent features that do not affect each other. The combined use of light intensity data and signal strength RSSI data can be more conducive to the distinction of different directions indoors, and effectively solve the positioning problem caused by high-frequency personnel flow and multiple regional similarities; other CSI channel features and signal strength RSSI data have feature redundancy to a certain extent. In addition, the material and human costs of collecting channel features are increased compared to the indoor light intensity information measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the adaptive stereo array indoor positioning method; Figure 2 It is a schematic diagram of an adaptive stereo array indoor positioning system; Figure 3 This is the Lo_RES network structure diagram; Figure 4 This is the system module diagram. DETAILED DESCRIPTION

[0016] The technical solution of the present invention is now further described in conjunction with the accompanying drawings and embodiments.

[0017] Example 1: Figure 1 As shown, this embodiment proposes an adaptive stereo array indoor positioning method, which mainly includes the following steps: Step 1: Divide the indoor area to be positioned of size M×N into several square sub-areas according to the minimum interval to collect environmental data of the area to be positioned; the side length of the minimum interval is set to 8d according to the positioning accuracy and the side length of the stereo array in step 1, and the total number of square sub-areas is i, which is expressed as: ;

[0018] Step 2: Use a stereo array to synchronously collect the signal strength RSSI data and light intensity data of the square sub-area for subsequent positioning; the stereo array used in this step is a cube with an edge length of d, four light intensity testers are set at the four vertices of the two diagonals on the top surface of the cube, and a WiFi signal receiver is placed on each of the four side edges of the cube. The stereo array can synchronously collect light intensity data and signal strength RSSI data through the equipped light intensity tester and WiFi signal receiver. This step takes into account the special indoor environment. The signal strength RSSI data will be affected by the multipath effect and cause positioning errors. Therefore, the particularity of indoor positioning is fully considered, that is, the light intensity information in the indoor environment is sufficient, and the signal strength RSSI data and the light intensity data are combined. If too much environmental information is obtained, there is a complex redundant relationship between the signal strength RSSI data and other channel characteristics, and the cost of obtaining more environmental information is not worth the improvement in positioning accuracy. However, the light intensity data and signal strength RSSI data collected in this embodiment do not have a redundant relationship in positioning information.

[0019] Step 3: Combined with the indoor environment category of the indoor area to be located, divide the 24 hours of the day into multiple collection periods. In each collection period, collect the environmental data corresponding to each square sub-area. During the collection, place the three-dimensional array at the center of the square sub-area. Each collection period consists of multiple collection cycles, and the collection cycles of different collection periods may be different. In each collection cycle, the four light intensity testers each collect light intensity data four times to obtain the light intensity data set L. The specific formula for L is: ; ; in, represents the four light intensity data collected by the nth light intensity tester on the cubic array located in the i-th square sub-area in one collection cycle, It represents the light intensity data collected for the first time by the nth light intensity tester in one collection cycle, and so on.

[0020] In each collection cycle, the four WiFi signal receivers collect signal strength RSSI data four times each, and obtain the signal strength RSSI data set R, which is expressed as: ; ; in, Represents the four signal strength RSSI data collected by the nth WiFi signal receiver on the cubic array located in the i-th square sub-area in one collection cycle, Indicates the signal strength RSSI data collected for the first time by the nth WiFi signal receiver in a collection cycle.

[0021] The environmental database Fg is composed of light intensity data and signal strength RSSI data, which is expressed as: ; ; In this embodiment, the 24 hours a day are divided into 5 different time periods: , , , and ; Time period The time period is 7-12 Beijing time. The time period is 12:00-14:00 Beijing time. The time period is 14:00-18:00 Beijing time. The time period is from 18:00 to 22:00 Beijing time. From 22:00 Beijing time to 7:00 the next morning. According to the actual environment and characteristics of these five time periods, the WiFi signal strength and light intensity can be used for independent positioning or joint positioning. , Time period During these two time periods, the positioning requirements are relatively high, so the positioning method combining light intensity and WiFi signal strength is used during these two time periods. During the lunch break, the WiFi signal strength independent positioning method is used. In this embodiment, the use of the WiFi signal strength independent positioning method does not mean that light intensity data is not collected, but the proportion of light intensity data in the environmental data is reduced, thereby reducing the collection of light intensity data during this time period. Generally, during the overtime period, the positioning demand and the mobility of people in the indoor environment are weaker than during the daytime working period. Therefore, during this period, the independent positioning method of WiFi signal strength is used to reduce the collection of light intensity signals. This is usually the time for people to rest at home. The positioning demand during this period is generally at the lowest level of the day, and the light intensity in both indoor and outdoor environments is relatively weak. Therefore, using the independent positioning method based on WiFi signal strength during this period can reduce the collection of light intensity signals.

[0022] In view of the above situation, during the time period , Time period , Time period The environmental data is collected in a period of 0.2s. , Time period Environmental data is collected with a collection cycle of 0.5s. Through the above time-divided data collection and selection of corresponding positioning methods, low-power positioning can be achieved on the basis of meeting positioning requirements, thereby reducing the overall positioning energy consumption.

[0023] Step 4: Divide the environmental data collected in step 3 into training set, validation set and test set according to the proportion. The specific operations include: Divide the environmental data collected in step 3 into time periods, and then , Time period , Time period Three environmental data are randomly selected from the environmental data every second. The environmental data selected according to this rule constitute the training set. 75% of the environmental data are randomly selected from the remaining environmental data to constitute the validation set and 25% of the environmental data to constitute the test set.

[0024] The environmental data is grid-preprocessed, including: combining 16 signal strength RSSI data in an environmental data into a 4×4 image data, combining 16 light intensity data into a 4×4 image data, and normalizing the two 4×4 image data to convert the data values ​​into numbers between [0,1].

[0025] Step 5: Construct a Lo_RES network suitable for positioning scenarios; the data format of the Lo_RES network input is 4×4×2×1, where 4×4 represents the input image data, 2 represents the number of channels of the input image data, the light intensity data and the signal strength RSSI data each occupy one channel, and 1 represents the image format is grayscale.

[0026] like Figure 3 As shown, the Lo_RES network constructed in this embodiment is first a convolution layer with a convolution kernel size of 2×2, 6 convolution kernels, a stride of 1, and a padding of 0. A first pooling layer is connected after the convolution layer, and a residual block network layer is connected after the pooling layer. The residual block network layer is composed of three residual network blocks. The input data of the first two residual network blocks are directly added to their output data on the channel as the input data of the next residual network block. The output of the third residual network block is connected to the fully connected layer through the second pooling layer to output the corresponding probability values ​​of the five most likely square sub-areas where the target is located, and the center coordinates of the square sub-area with the largest probability value are taken as the final estimated position coordinates of the target. In this embodiment, the second pooling layer is a global average pooling layer.

[0027] Each residual network block first has a convolutional layer with a kernel size of 1×1, a number of kernels of 24, a stride of 0, and a padding of 0, and then passes through a batch normalization (BN) layer and an activation function, and then connects a convolutional layer with a kernel size of 1×1, a number of kernels of 24, a stride of 0, and a padding of 0 and a BN layer.

[0028] Step 6: When training the constructed Lo_RES network, the parameters to pay attention to mainly include the amount of training data in each batch during training, the parameters of each network layer (ω, b), the learning rate λ and the maximum number of iterations E. The amount of training data in each batch will affect the speed of network training, and the learning rate will affect the rate of gradient descent update of network back propagation. Modulating the network suitable for data training can make the classification and positioning effect of the network better. At the end of each training, the validation set is used to detect the effect of the trained network. When the positioning accuracy meets the standard, the test set is used for test analysis. Otherwise, the parameters need to be adjusted again and the Lo_RES network needs to be trained again.

[0029] The positioning accuracy of the Lo_RES network is tested based on the test set and the optimal network parameters. The loss function Ls is introduced to judge the positioning accuracy, which is expressed as: ; in, Represents the center coordinates of the output square sub-region, Indicates the actual location coordinates of the user. Represents the mean, s represents the total test set data volume, and the positioning effect of the current Lo_RES network is determined by the loss function. When the positioning accuracy meets the standard, the optimal network parameters are saved for real-time positioning in the online stage.

[0030] The Lo_RES network structure and optimal network parameters are burned into the chip on the stereo array. During online positioning, only the stereo array needs to be carried to complete the positioning in real time. Otherwise, the network parameters are readjusted to train the Lo_RES network until the Lo_RES network achieves the required positioning accuracy.

[0031] Step 8: Users with positioning needs carry the stereo array, which collects signal strength RSSI data and light intensity data according to different collection periods, and preliminarily processes the collected environmental data according to the data preprocessing method, and then inputs it into the Lo_RES network that meets the positioning accuracy requirements. The user's location is determined based on the output of the Lo_RES network to complete adaptive high-precision positioning.

[0032] Taking the indoor positioning of a teaching and research room of a university in Jiangsu Province as an example, the adaptive stereo array indoor positioning method proposed in this embodiment is further described. Figure 2 As shown in the figure, the light intensity data set L collected in the teaching and research room of a university in Jiangsu Province is as follows, in Lux: ; ; The data set R of the collected signal strength RSSI data is in dBm, specifically:

[0033]

[0034] Build as Figure 3 The Lo_RES network structure shown is specifically as follows: the input data is a 4×4×2 data graph, and after passing through the Lo_RES network, a 1×2 regression prediction position coordinate is output, and there are 6 groups of actual position and predicted position coordinate sampling groups, as shown in Table 1; Table 1 Comparison table of actual position and predicted position

[0035] It can be seen that the average Euclidean distance error between the actual position and the predicted position is 0.685m, which belongs to sub-meter positioning accuracy, indicating that the positioning method provided by the present invention has high positioning accuracy.

[0036] Example 2: Figure 4 As shown, this embodiment proposes an adaptive stereo array indoor positioning system, which mainly includes the following modules: 2×2 WiFi signal receiver modules are set on the four side edges of the three-dimensional array to receive the signal strength RSSI data of the area to be tested in real time; The 2×2 light intensity tester modules are arranged at the four corners of the top surface of the three-dimensional array. Specifically, the light intensity tester module of this embodiment is a circular light intensity tester, which is used to receive the light intensity data of the area to be tested in real time. The arrangement scheme of each module proposed in this embodiment can maximize the intensity difference and effectively complete the positioning task.

[0037] Data storage module: used to store the real-time environmental data collected by the 2×2 WiFi signal receiver module and the 2×2 light intensity tester module; AI computing module: including an AI network storage module and an AI network hyperparameter module, which are used to input the collected real-time status information into the AI ​​network module for high-precision position prediction and output the corresponding position prediction coordinate value; the AI ​​network storage module stores the Lo_RES network designed in Example 1, which is used to provide a network framework for position prediction; the AI ​​network hyperparameter module stores the trained network parameters; I / O port module: used to adjust positioning accuracy according to requirements and update corresponding network hyperparameters.

Claims

1. An adaptive stereo array indoor positioning method, characterized in that: The following steps are involved: Divide the indoor area to be located into several square sub-areas, and collect environmental data of each square sub-area; the environmental data includes signal strength RSSI data and light intensity data; Preprocess the collected environmental data and construct training sets, validation sets, and test sets; The training set is used to train the constructed Lo_RES network. After each training, the validation set is used to detect the positioning accuracy of the Lo_RES network after training. The test set is used to test the trained Lo_RES network with the positioning accuracy reaching the standard, and finally the indoor positioning network is obtained. The indoor positioning network obtains environmental data to determine the square sub-area where it is located; The Lo_RES network includes: a first convolutional layer, a first residual network block, a second residual network block, a third residual network block and a fully connected layer; The output of the first convolutional layer is input to the first residual network block, the input and output of the first residual network block are added on the channel and input to the second residual network block, the input and output of the second residual network block are added on the channel and input to the third residual network block, the output of the third residual network block is connected to the fully connected layer after a global average pooling, the corresponding probability value of the square sub-region where the fully connected layer outputs is located, and the center coordinates of the square sub-region with the largest probability value are taken as the final inferred position information; Each residual network block consists of a second convolutional layer, a first batch normalization layer, an activation layer, a third convolutional layer, and a second batch normalization layer connected in sequence.

2. The adaptive stereo array indoor positioning method according to claim 1, characterized in that: The method of dividing the indoor area to be located into several square sub-areas and collecting environmental data of each square sub-area, wherein the environmental data includes signal strength RSSI data and light intensity data, can be replaced by the following steps: The indoor area to be located is divided into several square sub-areas, and the 24 hours of the day are divided into multiple collection time periods; in each collection time period, the environmental data corresponding to each square sub-area is collected.

3. The adaptive stereo array indoor positioning method according to claim 2, characterized in that: The environmental data corresponding to each square sub-area is collected in each collection period, specifically including: According to the environmental characteristics of each collection period, the proportion of signal strength RSSI data and light intensity data in the environmental data is adjusted to obtain the environmental data corresponding to each square sub-area.

4. The adaptive stereo array indoor positioning method according to claim 3, characterized in that: The preprocessing of the collected environmental data specifically includes: Combining the signal strength RSSI data into an image data, and performing normalization processing on the image data; The light intensity data is combined into another image data, and the image data is normalized.

5. The adaptive stereo array indoor positioning method according to claim 4, characterized in that: In the Lo_RES network, the input data of the first convolutional layer is represented as: ,in, Represents the input image data, 2 represents the number of channels of the input image data, and 1 represents the image format is grayscale.

6. The adaptive stereo array indoor positioning method according to claim 1, characterized in that: The first convolutional layer is a convolutional layer with a convolution kernel size of 2×2, a number of convolution kernels of 6, a stride of 1, and a padding of 0.

7. The adaptive stereo array indoor positioning method according to claim 1, characterized in that: The second convolution layer and the third convolution layer are both convolution layers with a convolution kernel size of 1×1, a number of convolution kernels of 24, a stride of 0, and a padding of 0.

8. An adaptive stereo array indoor positioning system, characterized in that: include: The three-dimensional array is used to obtain environmental data, input the environmental data into the indoor positioning network, and obtain its own location information; The indoor positioning network is trained according to the following steps: Divide the indoor area to be located into several square sub-areas, and collect environmental data of each square sub-area; the environmental data includes signal strength RSSI data and light intensity data; Preprocess the collected environmental data and construct training sets, validation sets, and test sets; The training set is used to train the constructed Lo_RES network. After each training, the validation set is used to detect the positioning accuracy of the Lo_RES network after training. The test set is used to test the trained Lo_RES network with the positioning accuracy reaching the standard, and finally the indoor positioning network is obtained. The Lo_RES network includes: a first convolutional layer, a first residual network block, a second residual network block, a third residual network block and a fully connected layer; the output of the first convolutional layer is input to the first residual network block, the input of the first residual network block and its output are added on the channel and input to the second residual network block, the input of the second residual network block and its output are added on the channel and input to the third residual network block, the output of the third residual network block is connected to the fully connected layer after a global average pooling, the corresponding probability value of the square sub-region where the fully connected layer outputs is located, and the center coordinates of the square sub-region with the largest probability value are taken as the final inferred position information; Each residual network block consists of a second convolutional layer, a first batch normalization layer, an activation layer, a third convolutional layer, and a second batch normalization layer connected in sequence.

9. The adaptive stereo array indoor positioning system according to claim 8, characterized in that: The stereo array comprises: WiFi signal receiver module, used to collect signal strength RSSI data in environmental data; Light intensity tester module, used to collect light intensity data from environmental data; The calculation module stores the indoor positioning network and obtains its own location information by using the signal strength RSSI data collected by the WiFi signal receiver module and the light intensity data collected by the light intensity tester module.

10. The adaptive stereo array indoor positioning system according to claim 8, characterized in that: The method of dividing the indoor area to be located into several square sub-areas and collecting environmental data of each square sub-area, wherein the environmental data includes signal strength RSSI data and light intensity data, can be replaced by the following steps: Divide the indoor area to be located into several square sub-areas, and divide the 24 hours of the day into multiple collection periods; According to the environmental characteristics of each collection period, the proportion of signal strength RSSI data and light intensity data in the environmental data is adjusted to obtain the environmental data corresponding to each square sub-area; The preprocessing of the collected environmental data specifically includes: combining the signal strength RSSI data into an image data, and normalizing the image data; combining the light intensity data into another image data, and normalizing the image data; In the Lo_RES network, the input data of the first convolutional layer is represented as: ,in, Represents the input image data, 2 represents the number of channels of the input image data, and 1 represents the image format is grayscale.

Images