A low-latency perception and recognition method for intelligent industrial doors

Abstract

The present invention relates to a low-latency perception and recognition method for intelligent industrial doors. The method comprises the following steps: installing a millimeter-wave radar above the intelligent industrial door to cover a triggering area; collecting radar echoes to generate a point cloud; recording a one-dimensional range image of an empty scene as a background frame; triggering the industrial door to open based on the motion point cloud; clustering static point clouds using the OPTICS algorithm; comparing the one-dimensional range image difference values ​​of the current frame and the background frame to assist in determining the existence of a target in the triggering area; and dynamically updating the background frame to avoid environmental interference. The method uses a single millimeter-wave radar sensor in combination with the OPTICS clustering algorithm to process the static point cloud, accurately identifies targets, avoids false alarms, and has the advantages of high-precision detection, high auxiliary judgment accuracy, and strong environmental adaptability. The method solves the problems of continuous switching, accidental closure, temporary storage of goods, and traffic jams in the prior art, and significantly improves detection accuracy and the operating efficiency of industrial doors.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial door automatic control, and in particular to a low-latency perception and recognition method for intelligent industrial doors. Background Art

[0002] With the rapid development of industrial automation and intelligence, industrial doors, as important channels for factory logistics, have a direct impact on production efficiency and safety through their automatic control technology. In recent years, industrial door detection technology has developed rapidly, with the introduction of a variety of sensor technologies to achieve functions such as automatic door opening and anti-smashing. For example, photoelectric sensors use the principle of light beam blocking to detect obstacles in front of the door, ultrasonic sensors determine the target distance by measuring the sound wave reflection time, and infrared sensors detect human activity based on changes in infrared reflection or thermal radiation.

[0003] Among them, Mozhibi's industrial door millimeter wave radar acceleration is advanced in the industry. Its products can not only provide trigger signals for opening doors, but also realize multiple functions such as anti-smashing, human and vehicle differentiation, and direction judgment. The realization of these functions enables industrial doors to automatically adapt to different traffic needs to a certain extent, improve the traffic efficiency of industrial doors, and effectively ensure the safety of pedestrians and vehicles entering and exiting. Although the existing industrial door detection technology has achieved certain results, there are still some problems and deficiencies in practical applications, as follows: (1) Continuous opening and closing problems. When a truck stops in the trigger area of ​​the industrial door radar and starts to unload, workers will frequently enter and exit the trigger area when carrying goods, causing the industrial door to open and close continuously. On the one hand, this accelerates the aging of the mechanical components of the industrial door and increases maintenance costs. On the other hand, frequent opening and closing operations can easily cause mechanical fatigue of the door body and reduce the service life of the industrial door. (2) The problem of mis-closing. In actual operation, when a truck is parked in the triggering area of ​​the industrial door radar, the driver may need to leave the cab briefly for work. At this time, the detection system of the industrial door cannot accurately judge the temporary nature of the driver's departure, and thus automatically closes the industrial door. After the driver returns, he needs to re-trigger the door lifting device or manually lift the door. This not only increases the tediousness of the operation, but may also cause the interruption of the production process and reduce production efficiency; (3) The problem of temporarily placing goods. During the loading and unloading process of goods, workers may not be able to transport all the goods at once due to manpower or equipment limitations, and choose to temporarily place some goods in the triggering area of ​​the industrial door radar. These temporarily placed goods may be misjudged as obstacles by the detection system, causing the industrial door to fall and close. The workers need to re-trigger and lift the industrial door before they can continue to carry out the goods. This undoubtedly increases the labor intensity and working time of the workers and affects the efficiency of cargo loading and unloading. Summary of the Invention

[0004] In view of this, the present invention provides a low-latency perception and recognition method for intelligent industrial doors, which can solve the problem that existing industrial doors rely on motion for target detection and cannot accurately identify static targets, resulting in continuous switching and incorrect closing, thereby improving the intelligence of industrial door detection as well as factory efficiency and safety.

[0005] To achieve the above objectives, the present invention provides a low-latency perception and recognition method for intelligent industrial doors, comprising the following steps:

[0006] S1. Use millimeter-wave radar to collect clutter point clouds and one-dimensional range images in the industrial door empty scene as historical one-dimensional range images;

[0007] S2. Obtain the original ADC signal of the millimeter-wave radar and preprocess it to obtain point cloud data;

[0008] S3. Determine whether there is a target in the millimeter-wave radar detection area based on the point cloud data, and control the state of the industrial door based on the judgment result;

[0009] S301. If a point cloud with a negative velocity toward the millimeter-wave radar exists in the current frame, the industrial door is triggered to open. If the velocity of the target point cloud continues to be negative, the industrial door remains open and all point cloud data of the current frame are always stored.

[0010] S302: If there is no point cloud with a negative velocity toward the millimeter-wave radar in the current frame, determine whether it is interference from a clutter point, including the following steps:

[0011] The static point clouds in the current frame within a radius of 0.1m with each point cloud in the historical one-dimensional range image as the center are considered as interference and removed;

[0012] Using the velocity value v of the moving point cloud of the previous frame of the static point cloud of the current frame as the radius, search for newly appeared static point clouds until all point clouds are traversed;

[0013] Use the OPTICS algorithm to cluster all static point clouds and extract the center points of the point cloud clusters. If a new static point cloud cluster exists, proceed to step S303.

[0014] S303: Taking the distance unit corresponding to the center of the point cloud cluster as the center, extract 50 distance units before and after the distance unit in the Y direction of the current frame's one-dimensional range image, a total of 100 distance units, as a comparison data segment, and compare it with the same position of the historical one-dimensional range image. If the difference exceeds a preset threshold, it is determined that the newly appeared static point cloud has a target in the detection area; otherwise, it is determined that there is no target in the detection area, and the state of the industrial door is automatically controlled based on the judgment result.

[0015] S304: Update the historical one-dimensional range image.

[0016] Preferably, the millimeter-wave radar is installed obliquely above the industrial door.

[0017] Preferably, the millimeter wave radar transmits electromagnetic waves through the transmitting antenna to cover the industrial door trigger area within one detection cycle, the receiving antenna receives the reflected echo data, and uses a mixer to combine the transmitting signal and the receiving signal to generate an intermediate frequency signal, and the intermediate frequency is sampled by the ADC module to obtain the original ADC data.

[0018] Preferably, the pretreatment comprises the following steps:

[0019] S101, performing one-dimensional FFT processing on AD sampling data of a single electromagnetic wave chirp emitted by the millimeter-wave radar to obtain a one-dimensional range image;

[0020] S102, performing Doppler FFT processing on the velocity dimension to obtain velocity information;

[0021] S103, performing incoherent accumulation processing to obtain a range-Doppler map;

[0022] S104, performing CFAR detection on the range-Doppler map to extract strong target points;

[0023] S105 , performing array angle measurement through multi-channel data to obtain the distance, speed, and scattering intensity of the target point, geometrically mapping the distance and angle information of the target point and outputting them in the form of a point cloud.

[0024] Preferably, the OPTICS algorithm performs clustering processing on all stationary point clouds to extract cluster center points, including the following steps:

[0025] Calculate the core distance of each point cloud as the minimum radius threshold required for the core object;

[0026] Determine the accessibility distance based on the density accessibility relationship between point clouds;

[0027] Based on the core distance and the reachability distance, a priority queue is established to sort the unprocessed point clouds in ascending order of the reachability distance;

[0028] Process the priority queue in sequence to extract the point cloud with the smallest reachability distance, and update the priority queue according to the density reachable points in the point cloud neighborhood until all point clouds are processed;

[0029] Different clusters are extracted based on the reachability distance distribution to generate clustering clusters.

[0030] Preferably, when there is no new static point cloud appearing in the detection area within 2 seconds, no target triggering, and the difference between the comparison data segment and the historical one-dimensional range image is lower than a set threshold, the historical one-dimensional range image is updated.

[0031] Preferably, the average value of the five one-dimensional range images after 2 seconds is calculated to update the historical one-dimensional range image, and the expression is:

[0032] .

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] By comprehensively analyzing the dynamic and static point cloud states, the present invention can reliably determine whether a target object exists within the trigger area, effectively reducing the risk of false detection and missed detection. This not only avoids unnecessary closing or opening of industrial doors due to false judgments, significantly improving the safety and reliability of industrial door operation, but also realizes intelligent control, thereby improving the passage efficiency of industrial doors. At the same time, the high-precision detection capability enables millimeter-wave radar to achieve accurate target detection without relying on complex algorithms.

[0035] After determining the target position based on the dynamic and static point clouds, the present invention uses the corresponding one-dimensional range image as an auxiliary judgment basis. The one-dimensional range image has an intelligent update function, which can prevent erroneous analysis caused by outdated data, greatly improving the accuracy of detecting the presence of targets in the trigger area, further enhancing the accuracy of detection, and effectively avoiding the occurrence of misjudgments and misjudgments.

[0036] The present invention controls the opening and closing timing of industrial doors by comprehensively judging the position, speed and other information of the target, effectively solving the problems of continuous opening and closing, accidental closing, temporary storage of goods and traffic jams in the prior art. It not only extends the service life of industrial doors and reduces maintenance costs, but also ensures smooth factory logistics and improves overall production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of the present invention;

[0038] Figure 2 A comparison curve diagram of the difference between the current frame and the historical frame and the preset threshold value of the present invention;

[0039] Figure 3 Schematic diagram of the location of new targets selected for dynamic and static point clouds in the present invention. DETAILED DESCRIPTION

[0040] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0041] This embodiment provides a low-latency perception and recognition method for intelligent industrial doors, including the following steps:

[0042] S1. Install the millimeter-wave radar at an angle above the industrial door to ensure that the electromagnetic waves emitted obliquely forward can cover the entire trigger area in front of the industrial door. Use the millimeter-wave radar to collect the clutter point cloud and one-dimensional range image in the empty scene of the industrial door as the historical one-dimensional range image.

[0043] This embodiment uses a single millimeter-wave radar sensor to cover detection from short distances (10 cm) to long distances (over 20 m). Regardless of ambient light conditions, or in fog, rain, or dust, it can reliably detect the distance, speed, and angle of fast-moving objects with high precision, minimizing false and missed detections. Thanks to its shorter wavelength and greater number of transmitting and receiving antennas, the millimeter-wave radar sensor can accurately detect the presence, movement speed, and position of multiple targets. It also provides multiple functions, such as multi-zone monitoring (for locating targets in the trigger area and tracking their movement).

[0044] S2. Obtain the original ADC signal of the millimeter-wave radar and preprocess it to obtain point cloud data;

[0045] During a detection cycle, the millimeter-wave radar transmits electromagnetic waves through the transmitting antenna to cover the trigger area of ​​the industrial door. The receiving antenna receives the reflected echo data, and a mixer is used to combine the transmitted and received signals to generate an intermediate frequency signal. The intermediate frequency is sampled by the ADC module to obtain the original ADC data.

[0046] Preprocessing the raw ADC data includes the following steps:

[0047] S101, performing one-dimensional FFT processing on AD sampling data of a single electromagnetic wave chirp emitted by the millimeter-wave radar to obtain a one-dimensional range image;

[0048] S102, performing Doppler FFT processing in the velocity dimension to obtain velocity information of the target;

[0049] S103. Since there are multiple receiving channels after Doppler-FFT, non-coherent accumulation processing is performed to obtain a Range-Doppler Map (RDM).

[0050] S104, performing CFAR detection on the range-Doppler map to extract strong target points;

[0051] S105 , performing array angle measurement through multi-channel data to obtain the distance, speed, and scattering intensity of the target point, geometrically mapping the distance and angle information of the target point and outputting them in the form of a point cloud.

[0052] S3. Determine whether there is a target in the millimeter-wave radar detection area based on the point cloud data, and control the state of the industrial door based on the judgment result;

[0053] S301: If there is a moving point cloud with a negative velocity toward the millimeter-wave radar in the current frame, the industrial door is triggered to open. If the velocity of the target point cloud continues to be negative, the industrial door remains open and all point cloud data of the current frame are always stored.

[0054] S302: If there is no moving point cloud with a negative velocity toward the millimeter-wave radar in the current frame, determining whether it is interference from clutter points includes the following steps:

[0055] With each point cloud in the historical one-dimensional range image as the center, the static point cloud in the current frame within a radius of 0.1m is considered as interference and removed to avoid interference with subsequent judgments. The speed of the static point cloud in the current frame is 0;

[0056] Use the velocity value v of the moving point cloud in the previous frame of the static point cloud of the current frame as the radius to search for newly appeared static point clouds until all point clouds are traversed;

[0057] The OPTICS algorithm is used to cluster all stationary point clouds and extract the center points of the point cloud clusters. As an efficient density-based clustering algorithm, the OPTICS algorithm has the advantages of being insensitive to parameters, able to discover clusters of different densities, and visualizing the results when processing stationary point cloud data, compared with the classic density-based clustering algorithm DBSCAN. This provides more reliable clustering results for subsequent point cloud data processing and analysis. The basic process of the OPTICS algorithm includes the following steps:

[0058] Calculate the core distance of each point in the data set. The core distance refers to the minimum radius centered on the point so that the number of its neighboring points reaches a preset threshold.

[0059] Initialize the reachability distance of all points to a sufficiently large value. The reachability distance is used to measure the distance from a point to the densely reachable points in its neighborhood. Initially, it is set to infinity or a value much larger than the data space range to ensure the correctness of subsequent updates.

[0060] Establish a priority queue to store unprocessed points and sort them in ascending order of reachability distance. Initially, all points can be added to the queue in a certain order (such as random order), but the order will be dynamically adjusted according to the reachability distance during actual processing.

[0061] During the processing, the point with the smallest reachability distance is taken from the priority queue and recorded as the current point. For the current point, all points in its neighborhood are determined, and the reachability distances of these neighboring points are updated based on the distribution of points in the neighborhood. Specifically, for each neighboring point, if the distance from the current point to the neighboring point is greater than the core distance of the current point, the reachability distance of the neighboring point remains unchanged. Otherwise, the reachability distance of the neighboring point is updated to the larger value of the core distance of the current point and the distance from the current point to the neighboring point. The updated neighboring point is reinserted into the priority queue for subsequent processing. The above steps are repeated until all points have been processed.

[0062] According to the distribution of accessibility distance, an ordered point sequence is generated, which reflects the density distribution and cluster structure in the dataset, laying the foundation for subsequent point cloud data processing and analysis;

[0063] The core idea of ​​the OPTICS algorithm is based on the concepts of density accessibility and density connectivity. It determines the density distribution structure in the data set by calculating the core distance and accessibility distance of each point.

[0064] In this embodiment, the OPTICS algorithm is used to cluster all stationary point clouds and extract the center points of the point cloud clusters, including the following steps:

[0065] Calculate the core distance of each point cloud as the minimum radius threshold required for the core object;

[0066] According to the density accessibility relationship between point clouds, the accessibility distance is determined to measure the distance between the point and the density-reachable points in the neighborhood;

[0067] Based on the core distance and reachability distance, a priority queue is established to sort the unprocessed point clouds in ascending order of reachability distance;

[0068] Process the priority queue in sequence to extract the point cloud with the smallest reachability distance, and update the priority queue according to the density reachable points in the point cloud neighborhood until all point clouds are processed;

[0069] Different clusters are extracted based on the reachability distance distribution to generate clusters. If a new stationary point cloud cluster exists, the process proceeds to step S303 . Cluster analysis of the stationary point cloud is performed to effectively identify regions and structures with different density characteristics in the point cloud, laying a solid foundation for subsequent point cloud data processing and analysis, and helping to better understand the characteristics of the scene or object represented by the point cloud.

[0070] S303: Taking the distance unit corresponding to the center of the point cloud cluster as the center, extract 50 distance units before and after the distance unit in the Y direction of the current frame's one-dimensional range image, a total of 100 distance units, as a comparison data segment, and compare it with the same position of the historical one-dimensional range image. If the difference exceeds a preset threshold, it is determined that the newly appeared static point cloud has a target in the detection area. Otherwise, it is determined that there is no target in the detection area, and the state of the industrial door is automatically controlled based on the judgment result.

[0071] S304: If there is no new static point cloud in the detection area within 2 seconds, no target triggering, and the difference between the comparison data segment and the historical one-dimensional range image is lower than the set threshold, the historical one-dimensional range image is updated. This is because the installation position of the millimeter-wave radar is fixed, and the clutter point cloud in the empty scene generally does not change. However, the environment changes suddenly, and the one-dimensional range image is extracted from the radar raw data cube, so the value has a large jump. Therefore, the one-dimensional range image needs to be updated to prevent the jump from being too large and exceeding the threshold.

[0072] The method for updating the historical one-dimensional range image is to calculate the average value of the five-frame one-dimensional range image after 2 seconds to update the historical one-dimensional range image. The average value of the five-frame one-dimensional range image plus the sum of the historical one-dimensional range image and itself is allowed to fluctuate no more than 3% to avoid misjudgment caused by a large difference with the one-dimensional range image of the empty scene. The update constraints are:

[0073] .

[0074] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A low-latency perception and recognition method for intelligent industrial doors, characterized in that: The following steps are involved: S1. Use millimeter-wave radar to collect clutter point clouds and one-dimensional range images in the industrial door empty scene as historical one-dimensional range images; S2. Obtain the original ADC signal of the millimeter-wave radar and preprocess it to obtain point cloud data; S3. Determine whether there is a target in the millimeter-wave radar detection area based on the point cloud data, and control the state of the industrial door based on the determination result; S301. If there is a moving point cloud with a negative velocity toward the millimeter-wave radar in the current frame, the industrial door is triggered to open. If the velocity of the target point cloud continues to be negative, the industrial door remains open and all point cloud data of the current frame are always stored. S302: If there is no moving point cloud with a negative velocity toward the millimeter-wave radar in the current frame, determining whether it is interference from clutter points includes the following steps: The static point clouds in the current frame within a radius of 0.1m with each point cloud in the historical one-dimensional range image as the center are considered as interference and removed; Using the velocity value v of the moving point cloud of the previous frame of the static point cloud of the current frame as the radius, search for newly appeared static point clouds until all point clouds are traversed; Use the OPTICS algorithm to cluster all static point clouds and extract the center points of the point cloud clusters. If a new static point cloud cluster exists, proceed to step S303. S303: Taking the distance unit corresponding to the center of the point cloud cluster as the center, extract 50 distance units before and after the distance unit in the Y direction of the current frame's one-dimensional range image, a total of 100 distance units, as a comparison data segment, and compare it with the same position of the historical one-dimensional range image. If the difference exceeds a preset threshold, it is determined that the newly appeared static point cloud has a target in the detection area; otherwise, it is determined that there is no target in the detection area, and the state of the industrial door is automatically controlled based on the judgment result. S304: Update the historical one-dimensional range image.

2. A low-latency perception and recognition method for intelligent industrial doors according to claim 1, characterized in that: The millimeter wave radar is installed obliquely above the industrial door.

3. The low-latency perception and recognition method for intelligent industrial doors according to claim 2 is characterized in that: During a detection cycle, the millimeter-wave radar transmits electromagnetic waves through the transmitting antenna to cover the industrial door trigger area, and the receiving antenna receives the reflected echo data. A mixer is used to combine the transmitted signal and the received signal to generate an intermediate frequency signal. The intermediate frequency is sampled by the ADC module to obtain the original ADC data.

4. The low-latency perception and recognition method for intelligent industrial doors according to claim 3 is characterized in that: The pretreatment comprises the following steps: S101, performing one-dimensional FFT processing on AD sampling data of a single electromagnetic wave chirp emitted by the millimeter-wave radar to obtain a one-dimensional range image; S102, performing Doppler FFT processing on the velocity dimension to obtain velocity information; S103, performing incoherent accumulation processing to obtain a range-Doppler map; S104, performing CFAR detection on the range-Doppler map to extract strong target points; S105 , performing array angle measurement through multi-channel data to obtain the distance, speed, and scattering intensity of the target point, geometrically mapping the distance and angle information of the target point and outputting them in the form of a point cloud.

5. The low-latency perception and recognition method for intelligent industrial doors according to claim 1 is characterized in that: The OPTICS algorithm clusters all stationary point clouds to extract cluster centers, including the following steps: Calculate the core distance of each point cloud as the minimum radius threshold required for the core object; Determine the accessibility distance based on the density accessibility relationship between point clouds; Based on the core distance and the reachability distance, a priority queue is established to sort the unprocessed point clouds in ascending order of the reachability distance; Process the priority queue in sequence to extract the point cloud with the smallest reachability distance, and update the priority queue according to the density reachable points in the point cloud neighborhood until all point clouds are processed; Different clusters are extracted based on the reachability distance distribution to generate clustering clusters.

6. The low-latency perception and recognition method for intelligent industrial doors according to claim 1 is characterized in that: If there is no new static point cloud in the detection area within 2 seconds, no target triggering, and the difference between the comparison data segment and the historical one-dimensional range image is lower than the set threshold, the historical one-dimensional range image is updated.

7. The low-latency perception and recognition method for intelligent industrial doors according to claim 6 is characterized in that: The average value of the five one-dimensional range images after 2 seconds is calculated to update the historical one-dimensional range image. The constraints are: 。

Images