Road slope measurement method and system based on mobile crowd sensing

Abstract

The application provides a road slope measurement method and system based on mobile crowd sensing, comprising the following steps: S1: dividing a to-be-predicted area and a predicted area, assigning a data collection task, and collecting data to the cloud; S2: after the cloud multi-modal sensing data is time-space aligned, performing frame-by-frame correlation matrix operation, fusing the coordinate and the attitude vector of the mobile device, and dividing the space position into a training set and a test set after removing data noise; S3: extracting time-domain features and frequency-domain features from the training set and the test set data, and calculating individual independent features; S4: training and testing the model by using the individual independent features, and further performing data aggregation and result optimization on the preliminary prediction result. The application is beneficial to efficiently and low-costly realizing large-scale road slope sensing, beneficial to establishing a city three-dimensional map, further enriching the content of the current digital map, and shortening the data updating cycle.

Description

Technical Field

[0001] This invention relates to the fields of mobile sensing and urban computing, specifically to a method and system for measuring road slope based on mobile crowd sensing. Background Technology

[0002] Geographic Information Systems (GIS) are widely used in people's production and daily life. Three-dimensional geographic information, such as natural terrain undulations and road slopes, plays a guiding role in tasks such as energy-saving route selection, road network construction, and natural disaster early warning. Traditional methods for measuring road slope mainly fall into two categories: satellite remote sensing, however, has low resolution; and manual ground surveying, which suffers from high measurement costs and long data update cycles. With the widespread adoption of the internet and mobile devices, mobile crowdsourcing sensing has become one of the means of road slope measurement, offering advantages such as low labor and time costs and timely data updates.

[0003] Mobile crowd sensing refers to using users' mobile terminal devices as sensing units to perceive surrounding environmental information, possessing basic computing and communication functions. Mobile crowd sensing can achieve task allocation and data collection via the internet, enabling large-scale data coverage and high-frequency updates. To date, some studies have employed mobile crowd sensing to measure road slope. These studies use vehicles such as cars that are stably moving on the road as mobile devices, employing either direct measurement methods based on a ground coordinate system or elevation angle calculation methods based on a vehicle coordinate system. However, these studies mainly suffer from the following three problems.

[0004] First, current algorithms all construct models based on steady-state measurement, requiring the mobile device to remain strictly stationary on the vehicle. However, vehicles do not move smoothly indefinitely; bumps and jolting inevitably cause the mobile device to vibrate relative to the vehicle body, directly affecting the measurement results. Current algorithms cannot compensate for this type of error.

[0005] Secondly, current measurement algorithms all use cars as the carrier, requiring mobile devices to be fixed to the car, and the experiment must ensure that the vehicle's direction of travel is parallel to the slope. Because of the stringent requirements on the carrier and the environment under test, this type of method is difficult to extend to rural roads and other roads that are difficult for cars to traverse, thus limiting the coverage of the data.

[0006] Third, the current algorithm uses Google elevation map data to correct data offsets, making it highly dependent on elevation data. Therefore, in areas not covered by Google elevation map or in areas with low accuracy, the algorithm's accuracy is greatly reduced.

[0007] Patent document CN100516773C discloses an on-board road slope angle measurement system and method. The on-board road slope angle measurement system includes: a GPS speed measuring instrument for measuring the vertical velocity Vv and horizontal velocity Vh of a moving vehicle at a given instant; a front suspension displacement sensor installed on the front suspension of the vehicle for measuring the displacement Zf of the vehicle body at the front suspension position perpendicular to the road plane at the given instant; a rear suspension displacement sensor installed on the rear suspension of the vehicle for measuring the displacement Zr of the vehicle body at the rear suspension position perpendicular to the road plane at the given instant; and an electronic control unit electrically connected to the GPS speed measuring instrument, the front suspension displacement sensor, and the rear suspension displacement sensor to obtain the values ​​of Vv, Vh, Zf, and Zr, and calculate the road slope angle θ according to a formula. However, this invention does not utilize user-provided baseline motion data to calculate individual-independent features, and does not use machine learning regression algorithms to train and test the model. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the purpose of this invention is to provide a road slope measurement method and system based on mobile crowd sensing.

[0009] A road slope measurement method based on mobile crowd sensing provided by the present invention includes:

[0010] Step S1: Divide the area to be predicted into the area that has been predicted, assign data collection tasks, and aggregate the data to the cloud;

[0011] Step S2: After spatiotemporal alignment, the cloud-based multimodal sensing data is subjected to frame-by-frame correlation matrix operations and fused into the coordinates and attitude vectors of the mobile device. After removing data noise, it is divided into training set and test set according to spatial location.

[0012] Step S3: Extract time-domain and frequency-domain features from both the training and test sets, and calculate individual-independent features;

[0013] Step S4: Train and test the model using individual-independent features, and further aggregate and optimize the preliminary prediction results.

[0014] Preferably, in step S1:

[0015] The task publishing platform divides the area to be predicted and the area already predicted based on the existing slope data and its measurement time, and uses the task publishing platform to assign data collection tasks to mobile device owners.

[0016] The predicted region meets the following criteria: the current region already has slope data and its information period is less than a preset threshold, thus it is a region with valid slope data; the data measured in the predicted region will be divided into training set data;

[0017] The region to be predicted meets the following conditions: the current region does not contain slope information or the information period of its data is greater than a preset threshold; the data measured in the region to be predicted is divided into test set data;

[0018] The data collection task treats the bicycle as a slope measurement tool, requiring the task performer to collect sensor data during the ride, and the experiment does not strictly restrict the placement of the mobile phone; the data collected by any user includes accelerometer, magnetometer, gyroscope and GPS positioning data; the platform will additionally require task performers who meet the preset conditions to turn on the camera and take pictures of the road conditions.

[0019] Preferably, in step S2:

[0020] Step S2.1: Perform spatiotemporal alignment on all data. All user-uploaded inertial sensor data are resampled at the same sampling rate to ensure that the inertial sensor data is aligned in the temporal domain. GPS information is used to spatially align the inertial sensor data of different users at the same location, and camera data is used to label the slope categories on the map.

[0021] Step S2.2: Using spatiotemporally aligned magnetometer, accelerometer, and gyroscope data, the motion of the mobile device in Earth coordinates is characterized, and the calculation results are represented by the correlation matrix between the Earth coordinate system and the mobile device's own coordinate system.

[0022] Step S2.3: Use Kalman filtering to remove noise from the correlation matrix sequence and restore the data; the calculation process relies on the iterative update of confidence weights and error variance, and the weights and variances are updated using the calculation results from previous time steps;

[0023] Step S2.4: Using the Kalman filtered data, segment the data according to the GPS positioning; for the data in the predicted area, divide it into the training set, and divide the rest into the test set; use the existing slope values ​​to label the training set data, and determine the uphill and downhill categories according to the user's GPS positioning change trend to complete the labeling of the training data.

[0024] Preferably, in step S3:

[0025] Calculate individual-independent features using user-provided baseline motion data;

[0026] The time-domain features include: kurtosis, skewness, mean, variance and mode, maximum value, minimum value and data range;

[0027] The frequency domain characteristics include: frequency domain DC component, frequency domain variance, spectral peak value, frequency domain energy, spectral kurtosis, and skewness;

[0028] The reference motion data is derived from the data provided by each user and meets the following conditions: a) the frequency domain components do not change within a preset time period; b) the signal oscillation amplitude stability meets the preset value.

[0029] The baseline motion data records the user's cycling behavior on level road sections where stability meets preset values. The data obtained after calculating the time and frequency domain correlations of any data segment are individual-independent features.

[0030] Preferably, in step S4:

[0031] By leveraging the individual-independent characteristics of the data, a machine learning regression algorithm is used to train and test the model. The model input is the characteristics of the signal within the observation window, and the output is the average slope within the time window. Taking advantage of the continuous time series of the dataset and the discrete distribution of the test sites, as well as the repetitive spatial distribution of the data, the preliminary prediction results are further aggregated and optimized.

[0032] The data aggregation utilizes multiple data points provided by different users at the same location, and after data filtering, aggregates all data using a mean merging method;

[0033] The optimization of the results leverages the complementary characteristics of long and short time windows in describing trend and detail changes to capture slope-related information. It uses prediction results corresponding to different window lengths, performs confidence analysis, and then aggregates them into the final slope measurement result.

[0034] A road slope measurement system based on mobile crowd sensing, provided by the present invention, includes:

[0035] Module M1: Divides the area to be predicted into the area that has been predicted, assigns data collection tasks, and aggregates the data to the cloud;

[0036] Module M2: After spatiotemporal alignment, the cloud-based multimodal sensing data is processed by frame-by-frame correlation matrix operations and fused into the coordinates and attitude vectors of the mobile device. After removing data noise, it is divided into training and test sets according to spatial location.

[0037] Module M3: Both the training and test sets are used to extract time-domain and frequency-domain features, and individual-independent features are calculated.

[0038] Module M4: Uses individual-independent features to train and test the model, and further aggregates and optimizes the preliminary prediction results.

[0039] Preferably, in module M1:

[0040] The task publishing platform divides the area to be predicted and the area already predicted based on the existing slope data and its measurement time, and uses the task publishing platform to assign data collection tasks to mobile device owners.

[0041] The predicted region meets the following criteria: the current region already has slope data and its information period is less than a preset threshold, thus it is a region with valid slope data; the data measured in the predicted region will be divided into training set data;

[0042] The region to be predicted meets the following conditions: the current region does not contain slope information or the information period of its data is greater than a preset threshold; the data measured in the region to be predicted is divided into test set data;

[0043] The data collection task treats the bicycle as a slope measurement tool, requiring the task performer to collect sensor data during the ride, and the experiment does not strictly restrict the placement of the mobile phone; the data collected by any user includes accelerometer, magnetometer, gyroscope and GPS positioning data; the platform will additionally require task performers who meet the preset conditions to turn on the camera and take pictures of the road conditions.

[0044] Preferably, in module M2:

[0045] Module M2.1: Performs spatiotemporal alignment on all data. All user-uploaded inertial sensor data are resampled at the same sampling rate to ensure that the inertial sensor data is aligned in the temporal domain. GPS information is used to spatially align the inertial sensor data of different users at the same location, and camera data is used to label the slope categories on the map.

[0046] Module M2.2: Using spatiotemporally aligned magnetometer, accelerometer, and gyroscope data, the motion of the mobile device in Earth coordinates is characterized, and the calculation results are represented using the correlation matrix between the Earth coordinate system and the mobile device's own coordinate system.

[0047] Module M2.3: Uses Kalman filtering to remove noise from the correlation matrix sequence and restore the data; the calculation process relies on the iterative update of confidence weights and error variance, using the calculation results from previous time steps to update the weights and variance;

[0048] Module M2.4: Utilizes the Kalman-filtered data and segments the data based on GPS positioning; data within the predicted area is assigned to the training set, while the remaining data is assigned to the test set; the training set data labels are marked using existing slope values, and the uphill / downhill category is determined based on the user's GPS positioning trend to complete the labeling of the training data.

[0049] Preferably, in module M3:

[0050] Calculate individual-independent features using user-provided baseline motion data;

[0051] The time-domain features include: kurtosis, skewness, mean, variance and mode, maximum value, minimum value and data range;

[0052] The frequency domain characteristics include: frequency domain DC component, frequency domain variance, spectral peak value, frequency domain energy, spectral kurtosis, and skewness;

[0053] The reference motion data is derived from the data provided by each user and meets the following conditions: a) the frequency domain components do not change within a preset time period; b) the signal oscillation amplitude stability meets the preset value.

[0054] The baseline motion data records the user's cycling behavior on level road sections where stability meets preset values. The data obtained after calculating the time and frequency domain correlations of any data segment are individual-independent features.

[0055] Preferably, in module M4:

[0056] By leveraging the individual-independent characteristics of the data, a machine learning regression algorithm is used to train and test the model. The model input is the characteristics of the signal within the observation window, and the output is the average slope within the time window. Taking advantage of the continuous time series of the dataset and the discrete distribution of the test sites, as well as the repetitive spatial distribution of the data, the preliminary prediction results are further aggregated and optimized.

[0057] The data aggregation utilizes multiple data points provided by different users at the same location, and after data filtering, aggregates all data using a mean merging method;

[0058] The optimization of the results leverages the complementary characteristics of long and short time windows in describing trend and detail changes to capture slope-related information. It uses prediction results corresponding to different window lengths, performs confidence analysis, and then aggregates them into the final slope measurement result.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. This invention facilitates the efficient and low-cost realization of large-scale road slope perception, promotes the establishment of urban 3D maps, further enriches the content of current digital maps, and shortens the data update cycle;

[0061] 2. For existing urban road networks, this invention can effectively utilize cyclists on the road to measure and update the slope information of the road network;

[0062] 3. For a constantly updated road network, this invention can screen out newly added slope measurement points, thereby updating the structure of the road network. Attached Figure Description

[0063] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0064] Figure 1 This is an example diagram of the present invention.

[0065] Figure 2 This is an example diagram of a common method for users to perform a crowd perception task in step A of the present invention.

[0066] Wherein, (i) means the user performs the task by placing the mobile device in their pocket; (ii) means the user performs the task by placing the mobile device in their backpack; and (iii) means the user performs the task by mounting the mobile device on the handlebars and turning on the camera. Detailed Implementation

[0067] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0068] Example 1:

[0069] To overcome the high cost and long data update cycle of traditional measurement methods, and to overcome the dependence on stability in common mobile sensing methods, this invention designs a road slope measurement system based on mobile crowdsourcing sensing, using the inertial sensors and cameras built into mobile terminal devices. This system enables low-cost road slope measurement. Furthermore, the crowdsourcing method used in this invention aggregates data from numerous data collectors, achieving large-scale and long-term sensing. Crowdsourcing data fusion further improves the accuracy of the measurement results, completing the road slope sensing task. The executor of the crowdsourced data collection task designed in this invention is a cyclist on the road. The terminal sensors used in this invention are located in the mobile terminal devices carried by the cyclist, such as mobile phones, smart bracelets, and smartwatches. These terminal sensors include accelerometers, gyroscopes, magnetometers, GPS devices, and cameras. The sensors used are essential for conventional mobile terminal devices and can be used to record the movement state of the mobile device itself, its environment, and its location.

[0070] The technical solution adopted by this invention to solve its technical problem is: a crowd-sensing system for road slope measurement, specifically including the following steps:

[0071] Step A: The task publishing platform divides the area to be predicted and the area already predicted based on the existing slope data and measurement time. It then uses the task publishing platform to assign data collection tasks to mobile device owners and aggregates the data to the cloud.

[0072] Step B: After spatiotemporal alignment, the cloud-based multimodal sensing data undergoes frame-by-frame correlation matrix operations to fuse the coordinates and attitude vectors of the mobile device. After Kalman filtering to remove data noise, the data is then divided into training and test sets based on spatial location.

[0073] Step C: Both the training and test sets are subjected to time-domain and frequency-domain feature extraction, and individual-independent features are calculated using the benchmark motion data provided by the user.

[0074] Step D: Utilizing the features extracted in the preceding steps, this step trains and tests the model using machine learning regression algorithms such as SVM. The model input consists of the signal features within the observation window, and the output is the average slope within the time window. Taking advantage of the near-continuous time series of the dataset and the discrete distribution of test sites, as well as the repetitive spatial distribution of the data, this step further aggregates and optimizes the preliminary prediction results.

[0075] During deployment, the aforementioned road slope measurement system executes steps ABCD cyclically to achieve periodic updates of experimental data and measurement results. Both the training and test sets required for the experiment are continuously updated, allowing the model to be trained using the latest training sets at any time.

[0076] In step A above, the predicted area meets the following condition: the current area already has slope data and its information period is less than a predetermined threshold, i.e., it has valid slope data. Valid slope data can originate from traditional measurement methods or other certified computational results. Generally, the data measured in the predicted area will be divided into training set data.

[0077] In step A above, the region to be predicted satisfies the following conditions: the current region does not contain slope information or the information period of its data is greater than a predetermined threshold. Generally, the data measured in the region to be predicted is divided into test set data.

[0078] Step A above describes a data collection task that treats the bicycle as a slope measurement tool. The task executor is required to collect various sensor data during the ride, and the experiment does not strictly restrict the placement of the mobile phone. Data collected by any user should include accelerometer, magnetometer, gyroscope, and GPS positioning data. Additionally, the platform will require task executors with the necessary equipment to use their cameras to photograph the road surface to determine road conditions and collect basic road data.

[0079] Step B above, according to the order of operations, includes the following steps:

[0080] Step B1: This step performs spatiotemporal alignment on all data. All user-uploaded inertial sensor data are resampled at the same sampling rate to ensure temporal alignment. GPS information is used to spatially align inertial sensor data from different users at the same location, and camera data is used to categorize slopes on the map.

[0081] Step B2: This step uses spatiotemporally aligned magnetometer, accelerometer, and gyroscope data to characterize the motion of the mobile device in Earth coordinates. The calculation results are represented using the correlation matrix between the Earth coordinate system and the mobile device's own coordinate system.

[0082] Step B3: This step uses Kalman filtering to remove noise from the correlation matrix sequence and restore the true data. Its calculation process relies on iterative updates of the confidence weights and data error variance, requiring updates to the weights and variance based on previous calculation results.

[0083] Step B4: This step utilizes the Kalman-filtered data to segment the data based on GPS positioning. Data within the predicted area is assigned to the training set, while the remaining data is assigned to the test set. Simultaneously, the training set data is labeled using existing slope values, and the uphill / downhill category is determined based on the user's GPS positioning trend, thus completing the labeling of the training data.

[0084] In step C above, the time-domain features include at least: 1) kurtosis 2) Skewness 3) Mean, variance, and mode; 4) Maximum, minimum, and data range.

[0085] In step C above, the frequency domain characteristics include at least: 1) frequency domain DC component; 2) frequency domain variance; 3) spectral peak value; 4) frequency domain energy; and 5) spectral kurtosis and skewness.

[0086] In step C above, the reference motion data originates from the data provided by each user and meets the following conditions: 1) the frequency domain components remain unchanged over a long period; 2) the signal oscillation amplitude is basically stable. It records the user's smooth riding behavior on a level road segment, and by calculating the time and frequency domain correlations with any provided data segment, individual-independent characteristics of the data can be obtained.

[0087] In step D above, the data aggregation utilizes multiple data points provided by different users at the same location. After data filtering, all data is aggregated using the mean merging method.

[0088] Step D above, the result optimization step, utilizes the complementary characteristics of long-time window and short-time window data in describing trend changes and detailed changes to capture slope length-related information. This step uses the prediction results corresponding to different window lengths, performs confidence analysis using frequency domain energy variance, and then aggregates them into the final slope measurement result.

[0089] Example 2:

[0090] Example 2 is a preferred example of Example 1, and is used to illustrate the present invention in more detail.

[0091] This invention provides a road slope measurement system based on mobile crowdsourcing sensing, which reflects road slope information by recording changes in human cycling behavior. The system captures changes in inertial sensor data during the user's cycling process and further extracts time-frequency domain features related to road slope using this sensor data. Then, the system aggregates numerous trajectories using the user's GPS positioning to update the two-dimensional spatial distribution of the road network, and integrates prediction results from multiple users to calculate road slope data, ultimately obtaining the three-dimensional spatial structure of the map. This invention incorporates multiple stages including data acquisition, data processing, and data computation, employing a learning framework combining digital signal processing and machine learning. This framework effectively learns the time-frequency correlation between road slope and sensor data, and applies the learned knowledge to road slope measurement and periodic data updates.

[0092] like Figure 1-2As shown, this invention provides a road slope measurement system based on mobile crowd sensing. Specifically, this embodiment includes the following steps:

[0093] Step A: The task publishing platform divides the area to be predicted and the area already predicted based on the existing slope data and measurement time. It then uses the task publishing platform to assign data collection tasks to mobile device owners and aggregates the data to the cloud.

[0094] Step B: After spatiotemporal alignment, the cloud-based multimodal sensing data undergoes frame-by-frame correlation matrix operations to fuse the coordinates and attitude vectors of the mobile device. After Kalman filtering to remove data noise, the data is then divided into training and test sets based on spatial location.

[0095] Step C: Both the training and test sets are subjected to time-domain and frequency-domain feature extraction, and individual-independent features are calculated using the benchmark motion data provided by the user.

[0096] Step D: Utilizing the features extracted in the preceding steps, this step uses the SVM regression algorithm to train and test the model, and performs preliminary predictions, data aggregation, and result optimization. The model input consists of the features of the signal within the observation window, and the output is the average slope within the time window.

[0097] The aforementioned road slope measurement system is deployed by executing steps ABCD in a cyclical manner. The training and test sets required for the experiment are continuously updated, allowing the model to be trained using the latest training sets at any time.

[0098] In step A above, the predicted area meets the following criteria: the current area already has slope data and its information period is less than a predetermined threshold, i.e., it is an area with valid slope data. Valid slope data can originate from traditional measurement methods or other certified computational results. Generally, the data measured in the predicted area will be divided into training set data.

[0099] In step A above, the region to be predicted satisfies the following conditions: the current region does not contain slope information or the information period of its data is greater than a predetermined threshold. Generally, the data measured in the region to be predicted is divided into test set data.

[0100] Step A above describes a data collection task that treats the bicycle as a slope measurement tool. Task participants are required to collect various sensor data while riding, and the experiment does not strictly restrict the placement of mobile phones; they are generally placed in pockets, trouser pockets, or backpacks. Data collected by any user should include accelerometer, magnetometer, gyroscope, and GPS positioning data. Additionally, the platform will require participants with the necessary equipment to use their cameras to photograph road conditions, used to determine road surface direction and collect basic road condition data. To balance data distribution, the platform will implement a reward mechanism to encourage task participants to perform the task in different areas.

[0101] Step B above, according to the order of operations, includes the following steps:

[0102] Step B1: This step performs spatiotemporal alignment on all data. All user-uploaded inertial sensor data is resampled at a sampling rate of 50Hz to ensure temporal alignment. GPS information is used to spatially align inertial sensor data from different users at the same location. User-uploaded camera data uses a target recognition algorithm to determine the direction of motion of objects within the field of view, thereby identifying the user's location (uphill or downhill). The results are marked on a map.

[0103] Step B2: This step uses the spatiotemporally aligned magnetometer M. t Accelerometer A t Static calculation results of the Earth coordinate system incidence matrix Using a gyroscope W t Calculate the dynamic cumulative result Y of the coordinate system incidence matrix t .

[0104] Step B3: This step will use the static calculation results and dynamic cumulative result Y t The input is fed into a Kalman filter for noise removal. The weights are calculated as K = Σ0(Σ1 + Σ0). -1 , where Σ0 and Σ1 represent and Y t The corresponding variance. Using Obtain the final calculation result and replace the original static calculation result. Used to update Σ0.

[0105] Among them, X t The output sequence of the correlation matrix after Kalman filtering. For the use of magnetometer M t Accelerometer A t The estimated sequence of the correlation matrix is ​​denoted as the static calculation result, Y. t To use the gyroscope W tThe estimated coordinate system incidence matrix is ​​denoted as the dynamic calculation result;

[0106] Step B4: This step utilizes the Kalman-filtered data to segment the data based on GPS positioning. Data within the predicted area is assigned to the training set, while the remaining data is assigned to the test set. Simultaneously, the training set data is labeled using existing slope values, and the uphill / downhill category is determined based on the user's GPS positioning trend, thus completing the labeling of the training data.

[0107] Step C above, where the time-domain features are calculated using a time-domain sequence X, includes at least: 1) kurtosis. 2) Skewness 3) Mean, variance, and mode; 4) Maximum, minimum, and data range.

[0108] Where E represents the calculation of the mathematical expectation of the random sequence, μ is the mean of the input sequence X, and σ is the variance of the sequence X;

[0109] In step C above, the frequency domain features are calculated using a time-frequency domain sequence S, and include at least: 1) frequency domain DC component; 2) frequency domain variance; 3) spectral peak value; and 4) frequency domain energy. 5) Spectral kurtosis Kurt(S) and skewness Skew(S).

[0110] Where M is the time-domain sequence length of the two-dimensional time-frequency domain sequence S, and N is the frequency-domain length of the two-dimensional time-frequency domain sequence S. ij Let i be the element in the frequency domain sequence S with time domain index i and frequency domain index j;

[0111] In step C above, the reference motion data originates from the data provided by each user and meets the following conditions: 1) the frequency domain components remain unchanged over a long period; 2) the signal oscillation amplitude is basically stable. This represents the user's stable riding data on a level road segment, and is related to the time domain of any provided data segment. Frequency domain related Then, individual-independent features of the data can be obtained.

[0112] Where m is the sequence time shift amount of the time-domain correlation operation, n is the vector index subscript, M is the sequence length, and X is the vector index. n+m Let X be any input data to be calculated. ref, As the baseline motion data, S n The result of FFT spectrum calculation for the input signal. The conjugate spectrum of the reference motion data;

[0113] In step D above, the data aggregation step utilizes multiple data points provided by different users at the same location and aggregates all data using a mean merging method.

[0114] Step D above, the result optimization step, utilizes the complementary characteristics of long-time window and short-time window data in describing trend changes and detailed changes to capture slope length-related information. This step uses the prediction results corresponding to different window lengths, performs confidence analysis using frequency domain energy variance, and then aggregates them into the final slope measurement result.

[0115] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0116] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for measuring road slope based on mobile crowd sensing, characterized in that, include: Step S1: Divide the area to be predicted into the area that has been predicted, assign data collection tasks, and aggregate the data to the cloud; Step S2: After spatiotemporal alignment, the cloud-based multimodal sensing data is subjected to frame-by-frame correlation matrix operations and fused into the coordinates and attitude vectors of the mobile device. After removing data noise, it is divided into training set and test set according to spatial location. Step S3: Extract time-domain and frequency-domain features from both the training and test sets, and calculate individual-independent features; Step S4: Train and test the model using individual-independent features, and further aggregate and optimize the preliminary prediction results.

2. The road slope measurement method based on mobile crowd sensing according to claim 1, characterized in that, In step S1: The task publishing platform divides the area to be predicted and the area already predicted based on the existing slope data and its measurement time, and uses the task publishing platform to assign data collection tasks to mobile device owners. The predicted region meets the following criteria: the current region already has slope data and its information period is less than a preset threshold, thus it is a region with valid slope data; the data measured in the predicted region will be divided into training set data; The region to be predicted meets the following conditions: the current region does not contain slope information or the information period of its data is greater than a preset threshold; the data measured in the region to be predicted is divided into test set data; The data collection task treats the bicycle as a slope measurement tool, requiring the task performer to collect sensor data during the ride, and the experiment does not strictly restrict the placement of the mobile phone; the data collected by any user includes accelerometer, magnetometer, gyroscope and GPS positioning data; the platform will additionally require task performers who meet the preset conditions to turn on the camera and take pictures of the road conditions.

3. The road slope measurement method based on mobile crowd sensing according to claim 1, characterized in that, In step S2: Step S2.1: Perform spatiotemporal alignment on all data. All user-uploaded inertial sensor data are resampled at the same sampling rate to ensure that the inertial sensor data is aligned in the temporal domain. GPS information is used to spatially align the inertial sensor data of different users at the same location, and camera data is used to label the slope categories on the map. Step S2.2: Using spatiotemporally aligned magnetometer, accelerometer, and gyroscope data, the motion of the mobile device in Earth coordinates is characterized, and the calculation results are represented by the correlation matrix between the Earth coordinate system and the mobile device's own coordinate system. Step S2.3: Use Kalman filtering to remove noise from the correlation matrix sequence and restore the data; The calculation process relies on the iterative update of confidence weights and error variance, using the calculation results from previous time steps to update the weights and variances; Step S2.4: Using the Kalman filtered data, segment the data according to the GPS positioning; for the data in the predicted area, divide it into the training set, and divide the rest into the test set; use the existing slope values ​​to label the training set data, and determine the uphill and downhill categories according to the user's GPS positioning change trend to complete the labeling of the training data.

4. The road slope measurement method based on mobile crowd sensing according to claim 1, characterized in that, In step S3: Calculate individual-independent features using user-provided baseline motion data; The time-domain features include: kurtosis, skewness, mean, variance and mode, maximum value, minimum value and data range; The frequency domain characteristics include: frequency domain DC component, frequency domain variance, spectral peak value, frequency domain energy, spectral kurtosis, and skewness; The reference motion data is derived from the data provided by each user and meets the following conditions: a) the frequency domain components do not change within a preset time period; b) the signal oscillation amplitude stability meets the preset value. The baseline motion data records the user's cycling behavior on level road sections where stability meets preset values. The data obtained after calculating the time and frequency domain correlations of any data segment are individual-independent features.

5. The road slope measurement method based on mobile crowd sensing according to claim 1, characterized in that, In step S4: By leveraging the individual-independent characteristics of the data, a machine learning regression algorithm is used to train and test the model. The model input is the characteristics of the signal within the observation window, and the output is the average slope within the time window. Taking advantage of the continuous time series of the dataset and the discrete distribution of the test sites, as well as the repetitive spatial distribution of the data, the preliminary prediction results are further aggregated and optimized. The data aggregation utilizes multiple data points provided by different users at the same location, and after data filtering, aggregates all data using a mean merging method; The optimization of the results utilizes the complementary characteristics of long and short time windows in describing trend changes and detailed changes to capture information related to slope length. It uses the prediction results corresponding to different window lengths, performs confidence analysis, and then aggregates them into the final slope measurement results.

6. A road slope measurement system based on mobile crowd sensing, characterized in that, include: Module M1: Divides the area to be predicted into the area that has been predicted, assigns data collection tasks, and aggregates the data to the cloud; Module M2: After spatiotemporal alignment, the cloud-based multimodal sensing data is processed by frame-by-frame correlation matrix operations and fused into the coordinates and attitude vectors of the mobile device. After removing data noise, it is divided into training and test sets according to spatial location. Module M3: Both the training and test sets are used to extract time-domain and frequency-domain features, and individual-independent features are calculated. Module M4: Uses individual-independent features to train and test the model, and further aggregates and optimizes the preliminary prediction results.

7. The road slope measurement system based on mobile crowd sensing according to claim 6, characterized in that, In module M1: The task publishing platform divides the area to be predicted and the area already predicted based on the existing slope data and its measurement time, and uses the task publishing platform to assign data collection tasks to mobile device owners. The predicted region meets the following criteria: the current region already has slope data and its information period is less than a preset threshold, thus it is a region with valid slope data; the data measured in the predicted region will be divided into training set data; The region to be predicted meets the following conditions: the current region does not contain slope information or the information period of its data is greater than a preset threshold; the data measured in the region to be predicted is divided into test set data; The data collection task treats the bicycle as a slope measurement tool, requiring the task performer to collect sensor data during the ride, and the experiment does not strictly restrict the placement of the mobile phone; the data collected by any user includes accelerometer, magnetometer, gyroscope and GPS positioning data; the platform will additionally require task performers who meet the preset conditions to turn on the camera and take pictures of the road conditions.

8. The road slope measurement system based on mobile crowd sensing according to claim 6, characterized in that, In module M2: Module M2.1: Performs spatiotemporal alignment on all data. All user-uploaded inertial sensor data are resampled at the same sampling rate to ensure that the inertial sensor data is aligned in the temporal domain. GPS information is used to spatially align the inertial sensor data of different users at the same location, and camera data is used to label the slope categories on the map. Module M2.2: Using spatiotemporally aligned magnetometer, accelerometer, and gyroscope data, the motion of the mobile device in Earth coordinates is characterized, and the calculation results are represented using the correlation matrix between the Earth coordinate system and the mobile device's own coordinate system. Module M2.3: Uses Kalman filtering to remove noise from the correlation matrix sequence and restore the data; The calculation process relies on the iterative update of confidence weights and error variance, using the calculation results from previous time steps to update the weights and variances; Module M2.4: Utilizes the Kalman-filtered data and segments the data based on GPS positioning; data within the predicted area is assigned to the training set, while the remaining data is assigned to the test set; the training set data labels are marked using existing slope values, and the uphill / downhill category is determined based on the user's GPS positioning trend to complete the labeling of the training data.

9. The road slope measurement system based on mobile crowd sensing according to claim 6, characterized in that, In module M3: Calculate individual-independent features using user-provided baseline motion data; The time-domain features include: kurtosis, skewness, mean, variance and mode, maximum value, minimum value and data range; The frequency domain characteristics include: frequency domain DC component, frequency domain variance, spectral peak value, frequency domain energy, spectral kurtosis, and skewness; The reference motion data is derived from the data provided by each user and meets the following conditions: a) the frequency domain components do not change within a preset time period; b) the signal oscillation amplitude stability meets the preset value. The baseline motion data records the user's cycling behavior on level road sections where stability meets preset values. The data obtained after calculating the time and frequency domain correlations of any data segment are individual-independent features.

10. The road slope measurement system based on mobile crowd sensing according to claim 6, characterized in that, In module M4: By leveraging the individual-independent characteristics of the data, a machine learning regression algorithm is used to train and test the model. The model input is the characteristics of the signal within the observation window, and the output is the average slope within the time window. Taking advantage of the continuous time series of the dataset and the discrete distribution of the test sites, as well as the repetitive spatial distribution of the data, the preliminary prediction results are further aggregated and optimized. The data aggregation utilizes multiple data points provided by different users at the same location, and after data filtering, aggregates all data using a mean merging method; The optimization of the results utilizes the complementary characteristics of long and short time windows in describing trend changes and detailed changes to capture information related to slope length. It uses the prediction results corresponding to different window lengths, performs confidence analysis, and then aggregates them into the final slope measurement results.

Images