Obstacle recognition methods, systems, devices, and engineering vehicles for use in engineering vehicles

Abstract

This application discloses an obstacle recognition method, system, device, and vehicle for engineering vehicles, relating to the field of vehicle blind spot monitoring. The method includes: determining a filtering threshold based on obtained dust concentration parameters, vibration amplitude parameters, and a preset reference threshold; suppressing noise in the obtained detection signal based on the filtering threshold, retaining effective signals with amplitudes not less than the filtering threshold; and identifying the obstacle type based on the obstacle's micro-motion characteristics of the effective signal. This application achieves adaptive matching between the filtering parameters and environmental interference intensity by dynamically determining the filtering threshold using dust concentration and vibration amplitude parameters. Simultaneously, it accurately distinguishes between living and non-living obstacles based on micro-motion characteristics. These two aspects synergistically improve the accuracy of obstacle recognition under varying interference intensity environments.

Description

Technical Field

[0001] This application relates to the field of vehicle blind spot monitoring technology, and in particular to an obstacle recognition method, system, device, and engineering vehicle for engineering vehicles. Background Technology

[0002] When engineering vehicles operate in scenarios such as mining and construction, multiple blind spots exist due to the obstruction of the vehicle body, leading to frequent collisions in these blind spots. These scenarios are often accompanied by harsh working conditions such as high dust levels and strong vibrations. Dust scattering increases clutter in the detection signal and blurs obstacle features, while vehicle vibration generates false obstacle point clouds and image shifts. The intensity of these two types of interference continuously changes with the operating conditions, causing the noise level of the blind spot monitoring signal to exhibit significant dynamic fluctuations.

[0003] However, existing blind spot monitoring solutions use fixed filtering parameters for noise suppression, which cannot adapt to dynamic changes in interference intensity. Specifically, when interference intensifies, the fixed filtering parameters are too low, leading to false alarms due to noise misjudgment. Conversely, when interference weakens, the fixed filtering parameters are too high, risking filtering out weak but valid obstacle signals, also resulting in a higher false alarm rate. Furthermore, existing solutions rely solely on macroscopic features such as distance, speed, or outline for obstacle identification. Inanimate objects also possess these macroscopic features, making it impossible to accurately distinguish between living and non-living obstacles. Therefore, existing blind spot monitoring solutions exhibit a mismatch between fixed filtering parameters and dynamic interference environments, and fail to effectively distinguish between living and non-living objects, leading to decreased obstacle recognition accuracy in blind spot monitoring solutions used on engineering vehicles. Summary of the Invention

[0004] In view of the above problems, this application provides an obstacle recognition method, system, device, and engineering vehicle for engineering vehicles, so as to improve the detection accuracy of obstacles in the blind spots of engineering vehicles. The specific solution is as follows:

[0005] The first aspect of this application provides an obstacle recognition method for engineering vehicles, comprising:

[0006] Obtain dust concentration parameters, vibration amplitude parameters, and detection signals;

[0007] Based on the dust concentration parameter, the vibration amplitude parameter, and the preset benchmark threshold, the filtering threshold is determined;

[0008] The detection signal is subjected to noise suppression processing based on the filtering threshold, which filters out noise signals with a signal amplitude less than the filtering threshold and retains valid signals with a signal amplitude not less than the filtering threshold.

[0009] The micro-motion features of the obstacle are extracted from the effective signal, and the type of obstacle is identified based on the micro-motion features.

[0010] In one possible implementation, determining the filtering threshold based on the dust concentration parameter, the vibration amplitude parameter, and a preset reference threshold includes:

[0011] The sum of the product of the dust concentration parameter and the first preset weighting coefficient corresponding to the dust concentration parameter, the product of the vibration amplitude parameter and the second preset weighting coefficient corresponding to the vibration amplitude parameter, and the preset benchmark threshold is determined as the filtering threshold.

[0012] In one possible implementation, prior to performing noise suppression processing on the probe signal based on the filtering threshold, the method further includes:

[0013] A static scene benchmark model is constructed based on the static environmental data of the detection area. The static scene benchmark model includes the signal features of static obstacles and a fixed background.

[0014] The detection signal is differentially canceled with the static scene reference model to filter out static clutter signals in the detection signal that are consistent with the time domain waveform of the static scene reference model, thereby obtaining the detection signal that includes dynamic obstacle signals that differ from the time domain waveform of the static scene reference model.

[0015] In one possible implementation, the micro-motion feature includes micro-motion frequency and displacement trajectory, and the identification of the type of obstacle based on the micro-motion feature includes:

[0016] The obstacles whose micro-motion frequency falls within a preset micro-motion frequency range of living organisms are selected. The preset micro-motion frequency range of living organisms refers to the low-frequency micro-motion characteristics calibrated by the living organism.

[0017] The displacement trajectory of the obstacle is identified, and if the displacement trajectory meets the preset life form trajectory identification conditions, the type of the obstacle is identified as a life form. The preset life form trajectory identification conditions are: the rate of change of the velocity of the displacement trajectory within a continuous time period is within a preset rate of change range, and the smoothness of the displacement trajectory is within a preset smoothness fluctuation range.

[0018] In one possible implementation, the detection signal includes a first sensing signal and a second sensing signal. The first sensing signal is acquired by a first sensing unit and includes a signal characterizing the micro-motion frequency of the obstacle. The second sensing signal includes a signal characterizing the displacement trajectory of the obstacle. Extracting the micro-motion features of the obstacle from the valid signal includes:

[0019] The first sensing signal and the second sensing signal are registered in coordinate system, and the micro-motion frequency of the obstacle is extracted from the registered first sensing signal, and the displacement trajectory of the obstacle is extracted from the registered second sensing signal.

[0020] In one possible implementation, prior to performing noise suppression processing on the probe signal based on the filtering threshold, the method further includes:

[0021] Based on the vibration amplitude parameters, a correction coefficient is dynamically generated, and based on the correction coefficient, vibration offset compensation is performed on the detected signal;

[0022] Based on the dust concentration parameter, the detection signal after vibration offset compensation is subjected to dust interference noise reduction processing, and the detection signal is subjected to adaptive noise reduction and contour feature enhancement.

[0023] In one possible implementation, the obstacle recognition method for engineering vehicles further includes:

[0024] The distance and motion state of the obstacle are determined based on the valid signal;

[0025] The collision risk level is determined by weighting the distance, the motion state, the dust concentration parameter, and the vibration amplitude parameter.

[0026] Output the type of obstacle and the warning control command corresponding to the collision risk level.

[0027] The second aspect of this application provides an obstacle recognition system for engineering vehicles, comprising:

[0028] The parameter acquisition module is used to obtain dust concentration parameters, vibration amplitude parameters, and detection signals.

[0029] The threshold determination module is used to determine the filtering threshold based on the dust concentration parameter, the vibration amplitude parameter, and the preset benchmark threshold;

[0030] The noise suppression module is used to perform noise suppression processing on the detection signal based on the filtering threshold, filtering out noise signals with a signal amplitude less than the filtering threshold, and retaining valid signals with a signal amplitude not less than the filtering threshold;

[0031] The identification module is used to extract the micro-motion features of the obstacle from the valid signal and identify the type of the obstacle based on the micro-motion features.

[0032] In one possible implementation, the threshold determination module is configured as follows:

[0033] The sum of the product of the dust concentration parameter and the first preset weighting coefficient corresponding to the dust concentration parameter, the product of the vibration amplitude parameter and the second preset weighting coefficient corresponding to the vibration amplitude parameter, and the preset benchmark threshold is determined as the filtering threshold.

[0034] In one possible implementation, the noise suppression module is further configured to:

[0035] A static scene benchmark model is constructed based on the static environmental data of the detection area. The static scene benchmark model includes the signal features of static obstacles and a fixed background.

[0036] The detection signal is differentially canceled with the static scene reference model to filter out static clutter signals in the detection signal that are consistent with the time domain waveform of the static scene reference model, thereby obtaining the detection signal that includes dynamic obstacle signals that differ from the time domain waveform of the static scene reference model.

[0037] In one possible implementation, the recognition module is configured to identify the type of obstacle based on the micro-motion features as follows:

[0038] The obstacles whose micro-motion frequencies fall within a preset range of micro-motion frequencies of living organisms are selected from the micro-motion features. The preset range of micro-motion frequencies of living organisms refers to low-frequency micro-motion features calibrated for living organisms. The micro-motion features include the micro-motion frequency and the displacement trajectory.

[0039] The displacement trajectory of the obstacle is identified, and if the displacement trajectory meets the preset life form trajectory identification conditions, the type of the obstacle is identified as a life form. The preset life form trajectory identification conditions are: the rate of change of the velocity of the displacement trajectory within a continuous time period is within a preset rate of change range, and the smoothness of the displacement trajectory is within a preset smoothness fluctuation range.

[0040] In one possible implementation, the detection signal includes a first sensing signal and a second sensing signal. The first sensing signal is acquired by a first sensing unit and includes a signal characterizing the micro-motion frequency of the obstacle. The second sensing signal includes a signal characterizing the displacement trajectory of the obstacle. The micro-motion features of the obstacle are extracted from the valid signal. When extracting the micro-motion features of the obstacle from the valid signal, the identification module is configured as follows:

[0041] The first sensing signal and the second sensing signal are registered in coordinate system, and the micro-motion frequency of the obstacle is extracted from the registered first sensing signal, and the displacement trajectory of the obstacle is extracted from the registered second sensing signal.

[0042] In one possible implementation, the noise suppression module is further configured to:

[0043] Based on the vibration amplitude parameters, a correction coefficient is dynamically generated, and based on the correction coefficient, vibration offset compensation is performed on the detected signal;

[0044] Based on the dust concentration parameter, the detection signal after vibration offset compensation is subjected to dust interference noise reduction processing, and the detection signal is subjected to adaptive noise reduction and contour feature enhancement.

[0045] In one possible implementation, the obstacle recognition system for engineering vehicles further includes:

[0046] The collision warning unit is used to determine the distance and motion state of the obstacle based on the effective signal; to determine the collision risk level by performing a weighted calculation based on the distance, the motion state, the dust concentration parameter, and the vibration amplitude parameter; and to output the type of the obstacle and the warning control command corresponding to the collision risk level.

[0047] The third aspect of this application provides an obstacle recognition device for engineering vehicles, including a processor and a memory;

[0048] The memory stores a computer program that, when executed by the obstacle recognition device, implements the obstacle recognition method for engineering vehicles as described in the first aspect and any implementation thereof.

[0049] The fourth aspect of this application provides an engineering vehicle, including: an obstacle recognition system for engineering vehicles as described in the second aspect of this application.

[0050] By employing the aforementioned technical solution, the obstacle recognition method, system, device, and engineering vehicle provided in this application, through configuring a dynamic filtering threshold based on dust concentration parameters, vibration amplitude parameters, and a preset benchmark threshold, replaces the fixed filtering parameters of existing technologies. This allows the filtering threshold to dynamically change with the interference intensity caused by dust scattering and vehicle vibration. When interference intensifies, the filtering threshold increases accordingly, effectively filtering out increased noise signals and preventing noise from being misjudged as valid signals, thus avoiding an increase in the false alarm rate. When interference weakens, the filtering threshold decreases accordingly, preventing the excessive filtering of weak valid obstacle signals, which could lead to an increase in the missed alarm rate. This resolves the compatibility contradiction between fixed filtering parameters and dynamic interference environments. Subsequently, by configuring the extraction of micro-motion features of obstacles from the noise-suppressed valid signals and identifying obstacle types based on these micro-motion features, the application utilizes the essential differences in micro-motion features between living and non-living objects, avoiding the risk of existing technologies that rely solely on macroscopic features such as distance, speed, and contour, failing to accurately distinguish between living and non-living obstacles. Therefore, this application improves the detection accuracy of obstacles within the blind spots of engineering vehicles. Attached Figure Description

[0051] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0052] Figure 1 A flowchart of an obstacle recognition method for engineering vehicles provided in this application;

[0053] Figure 2 A flowchart of a noise suppression process provided in this application;

[0054] Figure 3 A flowchart of a full-process obstacle recognition method provided in this application;

[0055] Figure 4 A block diagram of an obstacle recognition system for engineering vehicles provided in this application;

[0056] Figure 5 This application provides a structural schematic diagram of an obstacle recognition device for engineering vehicles. Detailed Implementation

[0057] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0058] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0059] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0060] The first aspect of this application provides an obstacle recognition method for engineering vehicles, the flowchart of which is shown below. Figure 1 As shown, the obstacle recognition method for engineering vehicles includes:

[0061] S101. Obtain dust concentration parameters, vibration amplitude parameters, and detection signals.

[0062] It should be noted that in practical applications, dust concentration and vibration amplitude parameters are quantitative parameters of environmental factors that interfere with the detection signal during obstacle recognition. Dust concentration characterizes the concentration of suspended particulate matter in the working environment, typically measured in micrograms per cubic meter (μg / m³). This concentration is collected in real-time by dust sensors installed on the engineering vehicle. The sampling range can be selected based on the application scenario of the engineering vehicle; for example, in a coal mine scenario, a dust sensor with a measurement accuracy covering 0 to 2000 μg / m³ can be used. Vibration amplitude characterizes the mechanical vibration intensity of the engineering vehicle or work platform, typically measured in millimeters. This amplitude is collected in real-time by vibration sensors installed on the engineering vehicle. The sampling range can be selected based on the vibration intensity of the engineering vehicle; for example, if the engineering vehicle is a 15-ton bulldozer, the appropriate vibration sensor's sampling range covers 0 to 20 mm, with a vibration frequency range of 5 to 50 Hz.

[0063] It should be noted that in practical applications, the detection signal refers to the raw signal collected by the sensing unit for obstacle detection, including millimeter-wave radar point cloud signals, visual image signals, etc. It should be understood that although this embodiment uses dust concentration parameters and vibration amplitude parameters as examples, in other embodiments, other parameters that may affect the quality of the detection signal, such as light intensity, temperature, and humidity, can be introduced as environmental interference parameters, as long as they can quantify the degree of interference of the environment on the detection signal.

[0064] In one possible implementation, the sensing unit for acquiring detection signals can be of various types, including at least one of millimeter-wave radar, camera, and lidar. This application does not impose excessive limitations on the specific type of sensing unit. Multiple sensing units can be selectively installed based on the blind spot range of the engineering vehicle. This application does not impose excessive limitations or elaborate on the specific installation location of the aforementioned sensing units.

[0065] It should be noted that, in practical application scenarios, the obstacle recognition method for engineering vehicles provided by the first aspect and any implementation thereof can be executed independently for each monitoring blind zone of the engineering vehicle, or it can be executed in conjunction with the detection signals of each monitoring blind zone.

[0066] S102. Determine the filtering threshold based on dust concentration parameters, vibration amplitude parameters, and preset benchmark thresholds.

[0067] It should be noted that in practical application scenarios, the aforementioned preset benchmark threshold can be a signal noise threshold value pre-calibrated in the factory state, which represents the minimum noise level in an ideal environment without dust or vibration. Its value can be calibrated based on the average noise floor value in an ideal environment.

[0068] It should be noted that in practical applications, the working environment of engineering vehicles is harsh, and the intensity of interference with the detection signal is dynamically changing. For example, when an engineering vehicle moves from a clean maintenance area into a high-dust working area, dust scattering increases clutter in the radar echo signal, requiring a higher filtering threshold to effectively filter out the clutter. Similarly, when the engineering vehicle vibrates more due to increased load (such as a bulldozer gathering piled coal), the vehicle shaking can cause false obstacle point clouds in the image signal or radar echo signal, necessitating an increased filtering threshold. If a fixed threshold is used in existing technology, setting the threshold too low will cause a large amount of noise to be misjudged as obstacles, while setting it too high will cause weak, valid obstacle signals to be filtered out. Therefore, this application determines the filtering threshold by configuring parameters based on dust concentration, vibration amplitude, and a preset reference threshold. This allows the dust concentration and vibration amplitude parameters to characterize the interference intensity of the current environment on the signal. Furthermore, the adjustment reference for the threshold value is determined based on the preset reference threshold, so that the obtained filtering threshold value matches the current environmental interference intensity. This ensures that the filtering threshold value can both guarantee noise suppression and retain the effective detection signal to the maximum extent.

[0069] S103. Based on the filtering threshold, noise suppression processing is performed on the detection signal to filter out noise signals with a signal amplitude less than the filtering threshold, and retain effective signals with a signal amplitude not less than the filtering threshold.

[0070] It should be noted that, in practical applications, the above-mentioned implementation method for noise suppression of the detection signal based on the filtering threshold may include the following: Figure 2 Steps S201 to S204 are shown below:

[0071] like Figure 2 The diagram shows a flowchart of a noise suppression process. The specific steps are as follows:

[0072] Step S201: Identify the signal amplitude of the detection signal collected by each sensing unit. Then trigger step S202.

[0073] Step S202: Determine whether the amplitude of the detected signal is less than the filtering threshold. If yes, then trigger step S203; otherwise, trigger step S204.

[0074] Step S203: Mark the detection signal as a noise signal and delete the detection signal.

[0075] Step S204: Output the detection signal as a valid signal.

[0076] In one possible implementation, the detection signal obtained in step S201 above can be range-Doppler spectrum data obtained by sampling, mixing and FFT transformation of millimeter-wave radar (usually FMCW system), with the processing granularity being a single range-velocity detection signal, rather than an entire time sequence signal.

[0077] In one possible implementation, the detection signal obtained in step S201 above can also be the average gray value of each region after dividing the image captured by the camera into regions, and the average gray value of each region is used as the detection signal and compared with the filtering threshold. If the average gray value is less than the filtering threshold, the gray image of that region is deleted and replaced with the background value or the smoothed gray value.

[0078] It should be noted that in practical application scenarios, since there can be various types of detection signals, the preset reference threshold and filter threshold can be set based on the above types to adapt to different types of detection signals.

[0079] It should be noted that this application performs noise suppression processing on the detection signal based on the filtering threshold, filters out noise signals with signal amplitudes less than the filtering threshold, and retains effective signals with signal amplitudes not less than the filtering threshold. This achieves the elimination of environmental noise introduced by dust and vibration, improves the accuracy of the effective signal in representing obstacle-related information, and thus improves the final obstacle recognition accuracy.

[0080] S104. Extract the micro-motion features of obstacles from the effective signals, and identify the type of obstacle based on the micro-motion features.

[0081] It should be noted that in practical applications, micro-motion characteristics refer to the minute motion features exhibited by the local components or the entire obstacle, beyond its macroscopic motion. Examples include low-frequency micro-motions unique to living organisms, such as chest rise and fall, and limb swaying. In practical applications, whether an obstacle is a living organism significantly impacts the output and execution of subsequent avoidance or collision avoidance commands. For instance, when a non-living obstacle is detected in the blind spot behind an engineering vehicle, only standard prompts (such as voice and light cues) are needed. However, if a living obstacle is detected, the driver must be immediately alerted, and further intervention (such as active speed reduction) is required. Therefore, this application improves the accuracy of obstacle type identification by configuring the extraction of micro-motion characteristics from valid signals and identifying the obstacle type based on these characteristics.

[0082] In one possible implementation, after identifying the type of obstacle, to facilitate further processing (such as early warning or data recording), redundant data related to the obstacle (such as motion state, position, and distance) can be redundant-eliminating and standardized. Specifically, for valid signals of obstacles whose types (living or non-living) have been identified, redundant data elimination, abnormal parameter correction, and deduplication processing are performed. Redundant data elimination removes duplicate detection data caused by overlapping detection from multiple sensing units; abnormal parameter correction corrects abnormal jumps in parameters such as distance and speed caused by transient interference; and deduplication merges redundant records generated by the same obstacle being detected multiple times in different valid signals. After the above redundancy elimination and standardization processing, the final output data includes the type, spatial coordinates, real-time distance, and motion state of obstacles within the blind zone.

[0083] In one possible implementation, after outputting data such as the type, spatial coordinates, real-time distance, and motion status of obstacles within the blind zone, a continuous obstacle tracking mechanism can be established by combining inter-frame obstacle association with Kalman filtering. This ensures stable and continuous output of effective data, providing reliable data support for subsequent risk decisions and early warning outputs. Specifically, inter-frame association matching is performed between consecutive frames by calculating the similarity of obstacle position, velocity, and contour features, associating the identification results of the same obstacle in different frames. Simultaneously, Kalman filtering is used to predict and smooth the obstacle's position, velocity, and motion direction, establishing a continuous tracking model for the obstacle's motion trajectory. This ensures that tracking continuity is maintained even when obstacles are briefly obstructed or the detection signal fluctuates.

[0084] This application replaces the fixed filtering parameters of existing technologies by configuring a dynamic filtering threshold based on dust concentration parameters, vibration amplitude parameters, and a preset benchmark threshold. This allows the filtering threshold to dynamically change with the interference intensity caused by dust scattering and vehicle vibration. When interference intensifies, the filtering threshold increases accordingly, effectively filtering out increased noise signals and preventing noise from being misjudged as valid signals, thus avoiding an increase in the false alarm rate. When interference weakens, the filtering threshold decreases accordingly, preventing the excessive filtering of weak valid obstacle signals, which would lead to an increase in the missed alarm rate. This resolves the compatibility contradiction between fixed filtering parameters and dynamic interference environments. Subsequently, by configuring the extraction of the micro-motion features of obstacles from the noise-suppressed valid signals, and identifying obstacle types based on these micro-motion features, this application utilizes the essential differences in micro-motion features between living and non-living objects, avoiding the risk of existing technologies that rely solely on macroscopic features such as distance, speed, and contour, failing to accurately distinguish between living and non-living obstacles. Therefore, this application improves the detection accuracy of obstacles within the blind spots of engineering vehicles.

[0085] In one possible implementation, the filtering threshold is determined based on dust concentration parameters, vibration amplitude parameters, and a preset reference threshold, including:

[0086] The sum of the product of the dust concentration parameter and the first preset weighting coefficient corresponding to the dust concentration parameter, the product of the vibration amplitude parameter and the second preset weighting coefficient corresponding to the vibration amplitude parameter, and the preset benchmark threshold is determined as the filtering threshold.

[0087] It should be noted that, in practical applications, the above filtering threshold T can be determined using the following calculation formula: Where C is the dust concentration parameter (unit: μg / m³). 3 k1 is the first preset weighting coefficient, A is the vibration amplitude parameter (unit: mm), k2 is the second preset weighting coefficient, T0 is the preset reference threshold, which is a dimensionless parameter calculated based on the noise floor amplitude of the detected signal, and the filter threshold T is also a dimensionless parameter.

[0088] It should be noted that, in practical applications, the specific values ​​of k1 and k2 can be determined by linearly calibrating the vibration interference intensity and noise amplitude. Specifically, in engineering vehicle operation scenarios, the noise amplitude generated by dust scattering and vehicle vibration has an approximately linear relationship with the interference intensity. The specific calibration steps are as follows: In an ideal environment with no dust and no vibration, the sensing unit continuously collects detection signals for 10 minutes, calculates the statistical average of the noise floor amplitude (which is dimensionless), and determines the statistical average as T0; then, keeping the vibration amplitude at 0, the dust concentration is gradually increased (0 to 2000 μg / m³) in a standard environmental chamber, and the noise floor increment ΔTC at different concentrations is collected. The least squares method is used to fit the linear relationship between ΔTC and C, and the slope is k1; keeping the dust concentration at 0, the vibration amplitude is gradually increased (0 to 20 mm) on a standard vibration table, and the noise floor increment ΔTA at different amplitudes is collected. The least squares method is used to fit the linear relationship between ΔTA and A, and the slope is k2. For example, in working environments with strong dust interference, k1 can be set to 0.005 m³ / μg; in working environments with strong vibration interference, the calibration value of k2 can be set to 0.3 / mm.

[0089] In one possible implementation, before performing noise suppression processing on the probe signal based on a filtering threshold, the following is also included:

[0090] A static scene baseline model is constructed based on the static environmental data of the detection area. The static scene baseline model includes the signal features of static obstacles and a fixed background.

[0091] The detection signal is differentially canceled with the static scene reference model to filter out static clutter signals in the detection signal that are consistent with the time domain waveform of the static scene reference model, and obtain a detection signal that includes dynamic obstacle signals that differ from the time domain waveform of the static scene reference model.

[0092] It should be noted that in the actual application scenarios of engineering vehicles, there are usually fixed obstacles such as walls, equipment, and material piles in the blind spots of engineering vehicles. The radar echo signals or visual contour features generated by these static obstacles are usually of high intensity and persistent. If they are directly used as the object of noise suppression processing, it will not only increase the computational burden, but may also lead to the risk of inaccurate noise suppression due to the reflection interference of fixed obstacles. Therefore, this application constructs a static scene benchmark model including static obstacles and fixed background signal features by configuring static environmental data based on the detection area, thereby realizing the accurate identification and marking of fixed obstacles.

[0093] It should be noted that in practical application scenarios, constructing a static scene baseline model based on static environmental data of the detection area can be achieved when the engineering vehicle is powered on or when there are no dynamic obstacles in the working environment. The sensing unit collects environmental data within the blind zone, and this environmental data characterizes the features of fixed obstacles in the current scene. These features include radar echo characteristics of fixed walls and visual contour characteristics of stationary equipment. Subsequently, these feature data of fixed obstacles are digitally modeled and stored using a pre-set model to construct the static scene baseline model. It can be understood that the aforementioned static scene baseline model can be a static background image of radar point clouds, a background frame of a visual image, or a fusion data model of both.

[0094] In one possible implementation, where the static scene baseline model is a static background image of a radar point cloud, the above implementation of constructing a static scene baseline model may include steps A1 to A10 below.

[0095] Step A1: Obtain raw point cloud data (including range, azimuth, elevation, and echo intensity information) from multiple frames of millimeter-wave radar acquisition. Then trigger step A2.

[0096] Step A2 involves using a point cloud preprocessing algorithm to filter out invalid points and downsample the voxel mesh. This triggers step A3.

[0097] In one possible implementation, the point cloud preprocessing algorithm described above may include an algorithm for filtering out invalid points whose distance is greater than a maximum distance threshold or whose distance is less than a minimum distance threshold, as well as a voxel grid filtering algorithm for downsampling.

[0098] Step A3 involves performing spatiotemporal registration of multi-frame point clouds using the ICP (Iterative Closest Point) algorithm, which then triggers step A4.

[0099] Step A4: Obtain the registered unified coordinate system point cloud dataset. Then trigger step A5.

[0100] Step A5: The point cloud space is divided into fixed-size voxels using a 3D spatial meshing algorithm. This triggers step A6.

[0101] Step A6: Obtain the statistical characteristics data (frequency of occurrence, average intensity, height standard deviation) for each voxel. Then trigger step A7.

[0102] Step A7: Use the Gaussian Mixture Model (GMM) algorithm to build a multi-Gaussian distribution model for each voxel. This triggers step A8.

[0103] Step A8: Obtain the background probability model parameters (mean μ, variance σ², weight w). Then trigger step A9.

[0104] Step A9: Perform ground point cloud segmentation using the RANSAC plane fitting algorithm. This triggers step A10.

[0105] Step A10: Obtain the static obstacle point cloud set and construct the radar static background map.

[0106] In one possible implementation, where the static scene reference model is the background frame of a visual image, the above-described implementation of constructing the static scene reference model may include steps B1 to B10 below.

[0107] Step B1: Obtain continuous video frame sequence data (RGB or HSV color space). Then trigger step B2.

[0108] Step B2 involves using image preprocessing algorithms to adjust resolution and perform Gaussian filtering for noise reduction. This triggers step B3.

[0109] Step B3: Obtain initialization parameters (Gaussian distribution number K = 3 to 5, learning rate α = 0.01). Then trigger step B4.

[0110] Step B4: Use the Gaussian Mixture Model (MOG2) algorithm to create a background model for each pixel. This triggers step B5.

[0111] Step B5: Obtain the K Gaussian distribution parameters (mean μ) for each pixel. k σ k 2 Weight w k This triggers step B6.

[0112] Step B6 involves dynamically updating the model parameters based on the new frame data using an online learning algorithm, which then triggers step B7.

[0113] Step B7: Obtain the top B distributions with the largest cumulative weights as the background distributions. Then trigger step B8.

[0114] Step B8: Generate a background frame image using a background extraction algorithm (taking the mean of the background distribution). This triggers step B9.

[0115] Step B9 involves using a morphological processing algorithm to perform an opening operation to remove noise, which then triggers step B10.

[0116] Step B10: Obtain the final visual background frame and complete the static scene baseline model.

[0117] It should be noted that in practical applications, the detection signal includes signals from dynamic obstacles, static obstacle signals from the static scene reference model, and fixed background signals. Therefore, this application configures the detection signal to undergo differential cancellation processing with the static scene reference model, filtering out static clutter signals in the detection signal that are consistent with the time-domain waveform of the static scene reference model. This yields a detection signal that includes dynamic obstacle signals that differ from the time-domain waveform of the static scene reference model. This approach effectively removes fixed background interference at the source of the detection signal, reducing the amount of data required for subsequent noise suppression and recognition processes, and improving the accuracy and efficiency of obstacle recognition.

[0118] It should be noted that in practical applications, differential cancellation processing can be obtained by performing differential operations on the detection signal and the static scene reference model. Specifically: for point cloud-type detection signals acquired by millimeter-wave radar, the difference between the real-time point cloud data (detection signal) and the point cloud data in the static scene reference model is calculated. For image-type detection signals acquired by a camera, the pixel (or grayscale) difference between the real-time image frame (detection signal) and the background frame in the static scene reference model is calculated. Since the time-domain waveforms of static obstacles are highly consistent in the detection signal and the static scene reference model, their signal amplitude will approach zero after the differential operation, thus being effectively filtered out. However, dynamic obstacles (such as walking people or moving vehicles) exist in the detection signal but not in the reference model, and significant difference signals will be retained after the differential operation. Through this differential cancellation processing, accurate removal of all static background clutter in the scene is achieved, improving the accuracy and intensity of the representation of dynamic obstacles in the detection signal.

[0119] In one possible implementation, the above differential cancellation processing can be implemented by independently encapsulating the impulse cancellation algorithm in the Moving Target Indication (MTI) algorithm to obtain the cancellation processing algorithm.

[0120] In one possible implementation, the micro-motion features include micro-motion frequency and displacement trajectory, and the type of obstacle is identified based on the micro-motion features, including:

[0121] The system filters out obstacles whose micro-motion frequencies fall within a preset range of micro-motion frequencies of living organisms. The preset range of micro-motion frequencies of living organisms refers to the low-frequency micro-motion characteristics calibrated by the living organisms.

[0122] The system identifies the displacement trajectory of obstacles and outputs the identification result that the obstacle type is a living being if the displacement trajectory meets the preset conditions for identifying living beings trajectory. The preset conditions for identifying living beings trajectory are: the rate of change of the velocity of the displacement trajectory within a continuous time period is within a preset rate of change range, and the smoothness of the displacement trajectory is within a preset smoothness fluctuation range.

[0123] It should be noted that in engineering vehicle operation scenarios, obstacles within blind spots include living beings (such as workers) and non-living obstacles (such as tarpaulins used to cover materials being blown by the wind). Non-living obstacles, due to their own movement or the influence of the external environment, will exhibit outlines and movements in radar echoes. If identification relies solely on distance, speed, or outline features, it may be impossible to accurately distinguish whether an obstacle is a living being, especially in dusty environments where it is invisible to the naked eye, thus hindering the driver's ability to make accurate collision avoidance decisions. Therefore, this application configures an obstacle type identification system based on micro-motion features, including micro-motion frequency and displacement trajectory, to accurately identify whether an obstacle is a living being, thereby assisting drivers in improving the accuracy of collision avoidance decisions.

[0124] It should be noted that, in practical applications, the aforementioned preset micro-motion frequency range for living organisms can be determined based on the organism's own movement frequency after calibration. For example, according to biological principles, when a human is at rest or walking slowly, the frequency of chest rise and fall (respiratory frequency) is typically between 0.1Hz and 2Hz, and the limb swing frequency also falls within this range. Therefore, the preset micro-motion frequency range for living organisms is set to 0.1Hz to 2Hz. This frequency range represents the unique low-frequency micro-motion characteristics of living organisms, and mechanical equipment, vibration interference, and the movement of debris do not possess stable micro-motion characteristics within this frequency range.

[0125] It should be noted that in practical applications, the micro-motion frequency can be obtained by performing time-frequency analysis on the effective signal retained after noise suppression, and then using the Fast Fourier Transform (FFT) algorithm or Wavelet Transform (WT) to extract the micro-motion frequency of the obstacle from the echo signal acquired by the millimeter-wave radar.

[0126] It should be noted that, in practical applications, the aforementioned preset rate of change range and preset smoothness fluctuation range can be determined based on calibration of human activity processes. Since the movement of living organisms is typically continuous, stable, and regular, while the movement trajectory of non-living entities (such as a cardboard box blown by the wind) is sudden, irregular, or discontinuous, this application identifies the displacement trajectory of an obstacle and calculates parameters such as the smoothness (i.e., jerk) and rate of change of velocity (i.e., acceleration) of the displacement trajectory. Then, by determining whether the rate of change of velocity of the displacement trajectory within a continuous time period falls within a preset rate of change range and whether the smoothness of the displacement trajectory falls within a preset smoothness fluctuation range, accurate detection of whether an obstacle is a living entity is achieved, reducing the risk of misjudgment.

[0127] In one possible implementation, the detection signal includes a first sensing signal and a second sensing signal. The first sensing signal is acquired by a first sensing unit and includes a signal characterizing the micro-motion frequency of the obstacle. The second sensing signal includes a signal characterizing the displacement trajectory of the obstacle. Extracting the micro-motion features of the obstacle from the valid signals includes:

[0128] The first and second sensing signals are registered in coordinate system. The micro-motion frequency of the obstacle is extracted from the registered first sensing signal, and the displacement trajectory of the obstacle is extracted from the registered second sensing signal.

[0129] It should be noted that in practical applications, the first sensing unit can be a millimeter-wave radar, and the second sensing unit can be a camera. Due to the physical limitations of a single sensing unit, it cannot collect relevant dimensional data, and its resistance to interference from different factors within the scene varies. For example, while millimeter-wave radar has advantages such as penetrating dust, resisting strong light, and high accuracy in speed and distance measurement, it lacks texture and contour information of obstacles, making it difficult to distinguish the movement patterns of inanimate obstacles with similar radar cross-sections. Conversely, while industrial cameras can provide rich contour and texture features, they are easily obstructed in high-dust and low-light environments, leading to obstacle loss. Therefore, this application achieves accurate acquisition of the required detection signals by configuring the first sensing unit to collect a first sensing signal including signals characterizing the micro-motion frequency of obstacles, and configuring the second sensing unit to collect a second sensing signal including signals characterizing the displacement trajectory of obstacles.

[0130] It should be noted that in practical applications, since the sensing signals collected by the first and second sensing units are signals of different dimensions, feature fusion requires coordinate system unification and feature registration. Specifically, millimeter-wave radar outputs distance-angle-velocity data in polar coordinates, while the camera outputs contour-texture data in image pixel coordinates; the two cannot be directly superimposed. Therefore, after obtaining the first and second sensing signals, a mapping matrix between the radar coordinate system and the visual image coordinate system can be established by calibrating the relative positions of the radar and camera. Subsequently, this mapping matrix is ​​used to project the target point cloud detected by the radar onto the visual image plane, completing coordinate system registration.

[0131] It should be noted that in practical applications, the extraction of the obstacle's micro-motion frequency from the registered first sensing signal and the extraction of the obstacle's displacement trajectory from the registered second sensing signal can be performed based on the signal characteristics of the corresponding sensing signals. For example, for the first sensing signal, the Fast Fourier Transform (FFT) algorithm is used to perform fast and slow time-dimensional transformations on the original radar echo signal to obtain the range-Doppler spectrum (including the radial velocity distribution of obstacles at various distances). Then, the Constant False Alarm Rate Detection (CFAR) algorithm is used to detect obstacles on the range-Doppler spectrum, extracting the slow-time-dimensional echo sequence corresponding to the obstacle (characterizing the echo intensity of the obstacle over a period of time). The Wavelet Threshold Denoising Algorithm is then used to suppress noise in the slow-time-dimensional echo sequence to obtain the denoised obstacle's micro-motion signal. Finally, the Fast Fourier Transform algorithm is used again to perform spectral analysis on the denoised micro-motion signal to obtain the obstacle's micro-motion frequency. For the second sensing signal, image recognition algorithms (such as YouOnly Look Once, YOLO) are used to detect targets in consecutive image frames, obtaining the bounding boxes and center pixel coordinates of obstacles in each frame (the bounding boxes represent the spatial range of obstacles in the image, and the center pixel coordinates represent the position of obstacles on the image plane). Then, the SORT (Simple Online and Realtime Tracking) algorithm is used to perform association matching processing on targets in consecutive frames to obtain obstacle tracking sequences with unique identifiers (representing the position change records of the same obstacle in consecutive frames, used to distinguish different obstacles). Furthermore, the coordinate system transformation matrix is ​​used to perform spatial mapping processing on the center pixel coordinates of obstacles to obtain the obstacle coordinate sequence in the world coordinate system (representing the absolute position change of obstacles in real three-dimensional space). Finally, the Kalman Filter Algorithm is used to smooth the obstacle coordinate sequence to obtain the continuous displacement trajectory of obstacles (representing the continuous motion path of obstacles in the time dimension).

[0132] In one possible implementation, before performing noise suppression processing on the probe signal based on a filtering threshold, the following is also included:

[0133] The correction coefficient is dynamically generated based on the vibration amplitude parameter, and the vibration offset compensation is performed on the detection signal based on the correction coefficient.

[0134] Based on the dust concentration parameter, the detection signal after vibration offset compensation is subjected to dust interference noise reduction processing, and the detection signal is subjected to adaptive noise reduction and contour feature enhancement.

[0135] It should be noted that in real-world applications, the severe vibrations of engineering vehicles not only generate noise but also cause physical shifts in the detection signals themselves. For example, the point cloud coordinates of millimeter-wave radar can drift due to vehicle vibrations, and camera image frames may appear misaligned or ghosted. Directly suppressing noise in such distorted signals risks misinterpreting valid target features as noise filtering or retaining incorrect coordinate information, leading to inaccurate subsequent identification. Furthermore, high-concentration dust environments not only generate clutter noise but also severely obstruct visual signals. Dust particles scatter light, causing images to appear washed out, with reduced contrast, and blurred or even lost target contours. Relying solely on filtering thresholds for noise suppression makes it difficult to recover obscured target features. Therefore, this application configures the dynamic generation of correction coefficients based on vibration amplitude parameters, performs vibration offset compensation on the detection signal based on the correction coefficients, and configures the dust concentration parameter to perform dust interference noise reduction processing on the detection signal after vibration offset compensation, and performs adaptive noise reduction and contour feature enhancement on the detection signal, thereby reducing the risk of noise suppression inaccuracy caused by the distortion of the detection signal due to vibration or dust obstruction, and improving the subsequent obstacle recognition accuracy.

[0136] It should be noted that, in practical applications, the above-mentioned implementation method of dynamically generating correction coefficients based on vibration amplitude parameters and compensating for vibration offset of the detected signal based on the correction coefficients can be as follows:

[0137] Input the vibration amplitude parameter A and the vibration frequency f into the fitting formula: Generate dimensionless correction coefficients ,in It is a dimensionless static calibration reference value. It is the amplitude correction factor (unit: 1 / mm). Characterizes the degree to which vibration amplitude affects signal offset. It is the frequency correction factor (unit: 1 / Hz). It characterizes the degree to which the vibration frequency affects the signal offset. , , The value can be determined based on calibration experiments of the radar and camera using a standard vibration table. Subsequently, based on the correction coefficient... Vibration correction is applied to the original radar point cloud coordinates (x, y, z) to obtain the corrected radar point cloud coordinates: ,in, , , This is the offset reference value in the corresponding vibration direction, pre-determined by calibration experiments, used to correct phase fluctuations in the echo signal caused by vibration. For the second sensing signal acquired by the camera, based on the correction coefficient... Affine Transformation Correction (ATC) is performed on image frames. This is done by calculating the offset vector between image frames caused by vibration and performing reverse compensation, thereby eliminating image misalignment caused by vibration.

[0138] It should be noted that, in practical applications, the above-mentioned method of dust interference denoising processing on the vibration offset compensated detection signal based on dust concentration parameters, and adaptive denoising and contour feature enhancement of the detection signal, can be implemented as follows: using a pixel-level adaptive denoising algorithm, denoising is performed based on a filtering window, wherein the size of the filtering window is adaptively adjusted according to the dust concentration parameter: when the dust concentration is low (e.g., below 500 μg / m³), the noise reduction is achieved by using a pixel-level adaptive denoising algorithm based on a filtering window. 3 When the dust concentration is moderate (e.g., 500 to 1000 μg / m³), a 3×3 filter window is used for mild noise reduction; when the dust concentration is moderate (e.g., 500 to 1000 μg / m³), a 3×3 filter window is used for mild noise reduction. 3 When the dust concentration is high (e.g., exceeding 1000 μg / m³), the filter window is expanded to 5×5 for moderate noise reduction; when the dust concentration is high (e.g., exceeding 1000 μg / m³), the filter window is expanded to 5×5 for moderate noise reduction. 3 When the filtering window is expanded to 7×7, depth denoising is performed. Then, the Sobel contour enhancement algorithm is used to calculate the horizontal gradient component G of the image. x and vertical gradient component G y And calculate the gradient magnitude. Then, the gradient magnitude is nonlinearly enhanced using a power-law transform G'=G γ (γ is a preset power value, γ>1, for example γ=1.5) The gradient magnitude G is nonlinearly enhanced to obtain the enhanced gradient magnitude G', which strengthens the contrast between the target edge and the background, allowing the blurred human outline and limb details obscured by dust to be restored and highlighted. Understandably, for millimeter-wave radar signals, the signal gain can also be adjusted according to the dust concentration parameter to compensate for the signal attenuation caused by dust scattering.

[0139] Furthermore, regarding the dust scattering clutter superposition problem in millimeter-wave radar, invalid clutter generated by dust particles can be filtered out by setting a point cloud distance threshold and a micro-motion feature threshold. Specifically: the point cloud distance threshold is used to limit the minimum detection distance of obstacles, and point cloud data with scattering below this distance threshold are judged as dust particle clutter and deleted. The micro-motion feature threshold is used to filter point cloud data with low-frequency micro-motion contour features of living organisms, retaining only point cloud data that simultaneously meets the distance threshold and micro-motion feature threshold conditions. This application filters out invalid clutter generated by dust particles by configuring and setting point cloud distance thresholds and micro-motion feature thresholds, thereby improving the accuracy of the output detection signal.

[0140] In one possible implementation, the obstacle recognition method for engineering vehicles provided in the first aspect of this application further includes:

[0141] Determine the distance and motion status of obstacles based on valid signals;

[0142] The collision risk level is determined by weighted calculation based on distance, motion state, dust concentration parameters, and vibration amplitude parameters.

[0143] Output the type of obstacle and the corresponding warning and control instructions based on the collision risk level.

[0144] It should be noted that in practical applications, the risk of collision with engineering vehicles varies significantly depending on the distance, motion state, type of obstacle, dust concentration, and vibration intensity. Different levels of collision risk necessitate corresponding safety measures. For example, a stationary person at a distance faces a lower collision risk in a low-dust environment, while a person approaching at close range faces an extremely high collision risk in a high-dust, high-vibration environment. Relying solely on a single parameter (such as obstacle distance) for risk assessment can easily lead to misjudgments or omissions. Therefore, this application determines the collision risk level by configuring a weighted calculation based on distance, motion state, dust concentration, and vibration amplitude parameters, taking into account the impact of current environmental factors on collision risk, thereby ensuring that the collision risk level is adapted to the collision risk warning requirements of the current environment.

[0145] In one possible implementation, the formula for the above weighted operation can be: R = Wd·fd(d) + Wv·fv(v) + Wc·fc(C) + Wa·fa(A), where R is the collision risk score (dimensionless, ranging from 0 to 1), d is the obstacle distance, v is the obstacle approaching velocity, C is the dust concentration parameter, A is the vibration amplitude parameter, fd(d) is the coefficient of influence of obstacle distance on collision risk, fv(v) is the coefficient of influence of obstacle approaching velocity on collision risk, fc(C) is the coefficient of influence of dust concentration parameter on collision risk, fa(A) is the coefficient of influence of vibration amplitude parameter on collision risk, and fd(d), fv(v), fc(C) and fa(A) are all dimensionless values ​​ranging from 0 to 1. fd is the distance risk function, which can be calculated using an inverse proportional function: fd(d) = d0 / (d + d0), where d0 is the distance reference constant, and the function value increases as the distance increases. fv is the velocity risk function, which can be calculated using a linear function: fv(v) = v / v0, where v0 is the velocity reference constant, and the function value increases as the velocity approaches the target velocity. fc is the dust risk function. As dust concentration increases, light attenuates due to dust-induced scattering, leading to lower visual recognition reliability at higher dust concentrations. Furthermore, the light attenuation rate varies across different dust concentration ranges due to differences in scattering types; for example, single scattering dominates at lower dust concentrations, while multiple scattering dominates at higher concentrations. Therefore, fc can be a piecewise linear function. This is achieved by calibrating the light attenuation rate across different dust concentration ranges and fitting the attenuation rate across the same dust concentration range to obtain the corresponding risk slope coefficient. A piecewise linear function is then constructed based on the risk slope coefficients for each dust concentration range. For example, when C is less than 500 μg / m³... 3 When the dust concentration is in the first dust concentration range, the corresponding function is: fc(C) = q1 × C + b1. When C is between 500 and 1000 μg / m³, 3 When the dust concentration is in the second dust concentration range, the corresponding function is: fc(C) = q² × C + b², where C is greater than 1000 μg / m³. 3 When the dust concentration reaches the third interval, the function corresponding to the third dust concentration interval is: fc(C) = q3×C + b3, where b1, b2, and b3 are the intercepts corresponding to the dust concentration intervals, used to limit the jumps in the output results and ensure that the piecewise linear function satisfies the continuity constraint. The segment points satisfy: q1×C + b1 = q2×C + b2, q2×C + b2 = q3×C + b3, where q1 is the risk slope coefficient corresponding to the first dust concentration interval, q2 is the risk slope coefficient corresponding to the second dust concentration interval, and q3 is the risk slope coefficient corresponding to the third dust concentration interval. The unit of the aforementioned risk slope coefficient is m³ / μg. fa is the vibration risk function. Since stronger vibrations result in lower radar point cloud stability, and the degree of impact on radar point cloud stability varies depending on whether the actual vibration exceeds the radar's vibration tolerance limit, for example, when the actual vibration is not greater than the vibration tolerance limit, the radar point cloud signal will exhibit signal blurring and point cloud drift. However, when the actual vibration exceeds the vibration tolerance limit, the radar point cloud signal will experience signal loss and point cloud fragmentation. Therefore, the aforementioned fa can also be a piecewise linear function. By calibrating the image offset rate of the sensing unit at each moment under different amplitude ranges, and fitting the image offset rates within the same amplitude range, the corresponding vibration risk function can be obtained. The risk slope coefficients corresponding to the amplitude intervals are used to construct piecewise linear functions based on these coefficients. For example, when A is in the first amplitude interval (less than 5mm), fa(A) = p1 × A + d1; when A is in the second amplitude interval (5mm to 15mm), fa(A) = p2 × A + d2; and when A is in the third amplitude interval (greater than 15mm), fa(A) = p3 × A + d3. Here, d1, d2, and d3 are the intercepts corresponding to the respective amplitude intervals, used to limit jumps in the output results and ensure the piecewise linear function satisfies continuity constraints. The segmentation points satisfy: p1 × A + d1 = p2 × A + d2, p2 × A + d2 = p3 × A + d3, where p1 is the risk slope coefficient corresponding to the first amplitude interval, p2 is the risk slope coefficient corresponding to the second amplitude interval, and p3 is the risk slope coefficient corresponding to the third amplitude interval. The unit of the aforementioned risk slope coefficient is 1 / mm. Wd, Wv, Wc, and Wa are the distance weight, velocity weight, dust weight, and vibration weight, respectively, all dimensionless values, and the sum of Wd, Wv, Wc, and Wa is 1. The values ​​of each weight are determined through statistical analysis of historical working data; for example, they can be 0.4, 0.3, 0.15, and 0.15, respectively. The comprehensive collision risk score R is obtained by weighted summation of the four factors. Based on preset thresholds, it is divided into three levels of collision risk: safe, warning, and dangerous. When R is less than the first preset threshold R1, it is a safe level; when R is not less than R1 and less than the second preset threshold R2, it is a warning level; and when R is not less than R2, it is a dangerous level. The specific values ​​of R1 and R2 can be set according to the risk tolerance of different working scenarios.

[0146] In one possible implementation, to facilitate data statistics and subsequent accident identification, this application may also log the data throughout the entire process and solidify and store the data after the engineering vehicle stops.

[0147] In one possible implementation, to avoid hardware or software failure, this application may also configure backup hardware and controllers and configure switching logic to automatically switch to backup hardware or software in the event of failure of the main device or main software, thereby improving the reliability of blind spot monitoring.

[0148] To facilitate understanding of the obstacle recognition method for engineering vehicles provided by the first aspect and any implementation thereof of this application, an example of a possible implementation of this application is described below:

[0149] like Figure 3 As shown, this is in the above Figure 1 The flowchart of a full-process obstacle recognition method based on this is provided below, and the specific operation steps are as follows:

[0150] Step S301: Obtain dust concentration parameters, vibration amplitude parameters, and detection signals. Then trigger step S302.

[0151] Step S302: Determine the filtering threshold based on dust concentration parameters, vibration amplitude parameters, and a preset benchmark threshold. Then, step S303 is triggered.

[0152] Step S303: Dynamically generate correction coefficients based on vibration amplitude parameters; perform vibration offset compensation on the detection signal based on the correction coefficients; perform dust interference noise reduction processing on the vibration offset compensated detection signal based on dust concentration parameters; and perform adaptive noise reduction and contour feature enhancement on the detection signal. This triggers step S304.

[0153] Step S304: Construct a static scene baseline model based on the static environmental data of the detection area. This triggers step S305.

[0154] Step S305: Perform differential cancellation processing on the detection signal and the static scene reference model to filter out static clutter signals in the detection signal that are consistent with the time domain waveform of the static scene reference model, and obtain a detection signal that includes dynamic obstacle signals that differ from the time domain waveform of the static scene reference model. Then, step S306 is triggered.

[0155] Step S306: Perform noise suppression processing on the probe signal based on the filtering threshold, filtering out noise signals with signal amplitudes smaller than the filtering threshold, and retaining valid signals with signal amplitudes not less than the filtering threshold. Then trigger step S307.

[0156] Step S307: Extract the micro-motion features of the obstacle from the valid signal, and identify the type of obstacle based on the micro-motion features. Then trigger step S308.

[0157] Step S308: Determine the distance and motion state of the obstacle based on the valid signal; perform weighted calculations based on the distance, motion state, dust concentration parameters, and vibration amplitude parameters to determine the collision risk level; and output the type of obstacle and the warning control command corresponding to the collision risk level.

[0158] The second aspect of this application provides an obstacle recognition system for engineering vehicles, the block diagram of which is shown below. Figure 4 As shown, it includes:

[0159] The parameter acquisition module 401 is used to acquire dust concentration parameters, vibration amplitude parameters, and detection signals;

[0160] The threshold determination module 402 is used to determine the filtering threshold based on dust concentration parameters, vibration amplitude parameters and preset reference thresholds;

[0161] The noise suppression module 403 is used to perform noise suppression processing on the detection signal based on the filtering threshold, filtering out noise signals with a signal amplitude less than the filtering threshold, and retaining effective signals with a signal amplitude not less than the filtering threshold.

[0162] The identification module 404 is used to extract the micro-motion features of obstacles from valid signals and identify the type of obstacles based on the micro-motion features.

[0163] In one possible implementation, the threshold determination module 402 is configured as follows:

[0164] The sum of the product of the dust concentration parameter and the first preset weighting coefficient corresponding to the dust concentration parameter, the product of the vibration amplitude parameter and the second preset weighting coefficient corresponding to the vibration amplitude parameter, and the preset benchmark threshold is determined as the filtering threshold.

[0165] In one possible implementation, the noise suppression module 403 is further configured to:

[0166] A static scene baseline model is constructed based on the static environmental data of the detection area. The static scene baseline model includes the signal features of static obstacles and a fixed background.

[0167] The detection signal is differentially canceled with the static scene reference model to filter out static clutter signals in the detection signal that are consistent with the time domain waveform of the static scene reference model, and obtain a detection signal that includes dynamic obstacle signals that differ from the time domain waveform of the static scene reference model.

[0168] In one possible implementation, the recognition module 404 is configured to: identify the type of obstacle based on micro-motion features.

[0169] The obstacle is selected from the micro-motion features whose micro-motion frequency falls within the preset micro-motion frequency range of the living organism. The preset micro-motion frequency range of the living organism is a low-frequency micro-motion feature calibrated by the living organism. The micro-motion features include micro-motion frequency and displacement trajectory.

[0170] The system identifies the displacement trajectory of obstacles and outputs the identification result that the obstacle type is a living being if the displacement trajectory meets the preset conditions for identifying living beings trajectory. The preset conditions for identifying living beings trajectory are: the rate of change of the velocity of the displacement trajectory within a continuous time period is within a preset rate of change range, and the smoothness of the displacement trajectory is within a preset smoothness fluctuation range.

[0171] In one possible implementation, the detection signal includes a first sensing signal and a second sensing signal. The first sensing signal is acquired by a first sensing unit and includes a signal characterizing the micro-motion frequency of the obstacle. The second sensing signal includes a signal characterizing the displacement trajectory of the obstacle. Micro-motion features of the obstacle are extracted from the valid signals. The identification module 404 is configured to:

[0172] The first and second sensing signals are registered in coordinate system. The micro-motion frequency of the obstacle is extracted from the registered first sensing signal, and the displacement trajectory of the obstacle is extracted from the registered second sensing signal.

[0173] In one possible implementation, the noise suppression module 403 is further configured to:

[0174] The correction coefficient is dynamically generated based on the vibration amplitude parameter, and the vibration offset compensation is performed on the detection signal based on the correction coefficient.

[0175] Based on the dust concentration parameter, the detection signal after vibration offset compensation is subjected to dust interference noise reduction processing, and the detection signal is subjected to adaptive noise reduction and contour feature enhancement.

[0176] In one possible implementation, the obstacle recognition system for engineering vehicles also includes:

[0177] The collision warning unit is used to determine the distance and motion status of obstacles based on valid signals; to determine the collision risk level by performing weighted calculations based on distance, motion status, dust concentration parameters, and vibration amplitude parameters; and to output the type of obstacle and the warning control command corresponding to the collision risk level.

[0178] The third aspect of this application provides an obstacle recognition device for engineering vehicles, including a processor and a memory;

[0179] The memory stores a computer program, which, when executed by the obstacle recognition device, implements an obstacle recognition method for engineering vehicles as described in the first aspect and any implementation thereof.

[0180] The fourth aspect of this application provides an engineering vehicle, including: an obstacle recognition system for engineering vehicles as described in the second aspect of this application.

[0181] The structural schematic diagram of the obstacle recognition device for engineering vehicles provided in the third aspect of this application can be seen as follows: Figure 5 As shown. The obstacle recognition device in this application embodiment may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 5 The obstacle recognition device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0182] like Figure 5 As shown, the obstacle recognition device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. When the obstacle recognition device is powered on, the RAM 503 also stores various programs and data required for the operation of the obstacle recognition device. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0183] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, memory cards, hard drives, etc.; and communication devices 509. Communication device 509 allows the obstacle recognition device to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An obstacle recognition device with various means is shown; however, it should be understood that it is not required to implement or possess all of the means shown. More or fewer means may be implemented or possessed alternatively.

[0184] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0185] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0186] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0187] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. An obstacle recognition method for engineering vehicles, characterized in that, include: Obtain dust concentration parameters, vibration amplitude parameters, and detection signals; Based on the dust concentration parameter, the vibration amplitude parameter, and the preset benchmark threshold, the filtering threshold is determined; The detection signal is subjected to noise suppression processing based on the filtering threshold, which filters out noise signals with a signal amplitude less than the filtering threshold and retains valid signals with a signal amplitude not less than the filtering threshold. The micro-motion features of the obstacle are extracted from the effective signal, and the type of obstacle is identified based on the micro-motion features.

2. The obstacle recognition method according to claim 1, characterized in that, The step of determining the filtering threshold based on the dust concentration parameter, the vibration amplitude parameter, and the preset benchmark threshold includes: The sum of the product of the dust concentration parameter and the first preset weighting coefficient corresponding to the dust concentration parameter, the product of the vibration amplitude parameter and the second preset weighting coefficient corresponding to the vibration amplitude parameter, and the preset benchmark threshold is determined as the filtering threshold.

3. The obstacle recognition method according to claim 1, characterized in that, Before performing noise suppression processing on the probe signal based on the filtering threshold, the method further includes: A static scene benchmark model is constructed based on the static environmental data of the detection area. The static scene benchmark model includes the signal features of static obstacles and a fixed background. The detection signal is differentially canceled with the static scene reference model to filter out static clutter signals in the detection signal that are consistent with the time domain waveform of the static scene reference model, thereby obtaining the detection signal that includes dynamic obstacle signals that differ from the time domain waveform of the static scene reference model.

4. The obstacle recognition method according to claim 1, characterized in that, The micro-motion features include micro-motion frequency and displacement trajectory, and the identification of the type of obstacle based on the micro-motion features includes: The obstacles whose micro-motion frequency falls within a preset micro-motion frequency range of living organisms are selected. The preset micro-motion frequency range of living organisms refers to the low-frequency micro-motion characteristics calibrated by the living organism. The displacement trajectory of the obstacle is identified, and if the displacement trajectory meets the preset life form trajectory identification conditions, the type of the obstacle is identified as a life form. The preset life form trajectory identification conditions are: the rate of change of the velocity of the displacement trajectory within a continuous time period is within a preset rate of change range, and the smoothness of the displacement trajectory is within a preset smoothness fluctuation range.

5. The obstacle recognition method according to claim 4, characterized in that, The detection signal includes a first sensing signal and a second sensing signal. The first sensing signal is acquired by a first sensing unit and includes a signal characterizing the micro-motion frequency of the obstacle. The second sensing signal includes a signal characterizing the displacement trajectory of the obstacle. Extracting the micro-motion features of the obstacle from the valid signals includes: The first sensing signal and the second sensing signal are registered in coordinate system, and the micro-motion frequency of the obstacle is extracted from the registered first sensing signal, and the displacement trajectory of the obstacle is extracted from the registered second sensing signal.

6. The obstacle recognition method according to claim 1, characterized in that, Before performing noise suppression processing on the probe signal based on the filtering threshold, the method further includes: Based on the vibration amplitude parameters, a correction coefficient is dynamically generated, and based on the correction coefficient, vibration offset compensation is performed on the detected signal; Based on the dust concentration parameter, the detection signal after vibration offset compensation is subjected to dust interference noise reduction processing, and the detection signal is subjected to adaptive noise reduction and contour feature enhancement.

7. The obstacle recognition method according to claim 1, characterized in that, The obstacle recognition method for engineering vehicles further includes: The distance and motion state of the obstacle are determined based on the valid signal; The collision risk level is determined by weighting the distance, the motion state, the dust concentration parameter, and the vibration amplitude parameter. Output the type of obstacle and the warning control command corresponding to the collision risk level.

8. An obstacle recognition system for engineering vehicles, characterized in that, include: The parameter acquisition module is used to obtain dust concentration parameters, vibration amplitude parameters, and detection signals. The threshold determination module is used to determine the filtering threshold based on the dust concentration parameter, the vibration amplitude parameter, and the preset benchmark threshold; The noise suppression module is used to perform noise suppression processing on the detection signal based on the filtering threshold, filtering out noise signals with a signal amplitude less than the filtering threshold, and retaining valid signals with a signal amplitude not less than the filtering threshold; The identification module is used to extract the micro-motion features of the obstacle from the valid signal and identify the type of the obstacle based on the micro-motion features.

9. An obstacle recognition device for engineering vehicles, characterized in that, Including processor and memory; The memory stores a computer program that, when executed by the obstacle recognition device, implements the obstacle recognition method for engineering vehicles as described in any one of claims 1 to 7.

10. An engineering vehicle, characterized in that, include: The obstacle recognition system for engineering vehicles as described in claim 8.

Images