A robot obstacle recognition method, device, equipment and medium

Abstract

The application discloses a kind of robot obstacle identification method, device, equipment and medium.Therein, the method includes: obtaining the echo data of multiple ultrasonic probes in the process of robot movement in set period and the real-time position information of robot;Determine target acoustic image according to the echo data and the real-time position information;When triggering identification processing condition, the target acoustic image is input into identification model according to fixed inference period, and initial obstacle identification result is obtained;Filter processing is carried out on the initial obstacle identification result of multiple inference periods, and target obstacle identification result is obtained.The technical scheme is based on the echo data of multiple ultrasonic probes to determine the corresponding acoustic image, and realizes the detection and identification of obstacle target in combination with identification model, can better identify the relevant information of obstacle target, improve the identification precision of obstacle target, and is not easily interfered by environment visibility.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to a method, apparatus, device, and medium for robot obstacle recognition. Background Technology

[0002] Obstacle detection and recognition are the core components for mobile robots to achieve safe obstacle avoidance and autonomous navigation during autonomous movement and operation.

[0003] At present, the mainstream approach in the industry is to use LiDAR and visual cameras to build a multimodal obstacle detection system, thereby combining the detection advantages of the two types of sensors to achieve environmental perception and target recognition. However, this traditional solution still has many defects and limitations in practical applications.

[0004] LiDAR boasts advantages such as high positioning accuracy and precise ranging, accurately reflecting the spatial position information of obstacles, making it one of the core components for robot environmental perception. However, this sensor suffers from problems such as signal saturation and blind spots in close-range detection, easily leading to missed detection of nearby obstacles and target loss. To adapt to different operating environments, repeated debugging of equipment parameters is required, which is challenging. Moreover, the hardware cost of LiDAR increases significantly with the improvement of detection accuracy, and different application scenarios often require the selection of different specifications of LiDAR, which not only increases the hardware design cost but also the subsequent equipment maintenance cost. LiDAR has weak detection capabilities for transparent and highly reflective objects, and its detection performance is significantly degraded in conditions with a lot of smoke and dust, resulting in poor environmental adaptability.

[0005] Visual cameras, relying on mature image recognition algorithms, can perform tasks such as obstacle classification and contour recognition, and have a comprehensive application system. However, pure vision solutions are highly dependent on lighting conditions, and their detection performance drops sharply in dark or low-light environments, requiring the use of other sensors for assistance. Even when using active infrared cameras, they are affected by the reflectivity of the object being measured and the visibility at the scene. Their recognition accuracy decreases in smoke and dust environments, and their effective detection distance is limited, greatly restricting the robot's operating scenarios. Summary of the Invention

[0006] This invention provides a robot obstacle recognition method, device, equipment, and medium. It determines the corresponding acoustic image based on the echo data of multiple ultrasonic probes and combines it with a recognition model to realize the detection and recognition of obstacle targets. It can better identify relevant information of obstacle targets, improve the recognition accuracy of obstacle targets, and is not easily affected by the visibility of the environment.

[0007] According to one aspect of the present invention, a robot obstacle recognition method is provided, wherein a plurality of ultrasonic probes are deployed on the robot according to preset installation conditions; the method includes: Acquire echo data from multiple ultrasonic probes during robot movement within a set period, as well as the robot's real-time position information; The target acoustic image is determined based on the echo data and the real-time location information; When the recognition processing condition is triggered, the target acoustic image is input into the recognition model according to a fixed inference cycle to obtain the initial obstacle recognition result. The initial obstacle recognition results from multiple inference cycles are filtered to obtain the target obstacle recognition results.

[0008] According to another aspect of the present invention, a robot obstacle recognition device is provided, wherein a plurality of ultrasonic probes are deployed on the robot according to preset installation conditions; the device includes: The data acquisition module is used to acquire echo data from multiple ultrasonic probes during the robot's movement within a set period, as well as the robot's real-time position information. The image determination module is used to determine the target acoustic image based on the echo data and the real-time location information; The obstacle recognition module is used to input the target acoustic image into the recognition model according to a fixed inference cycle when the recognition processing condition is triggered, so as to obtain the initial obstacle recognition result. The filtering module is used to filter the initial obstacle recognition results from multiple inference cycles to obtain the target obstacle recognition result.

[0009] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the robot obstacle recognition method according to any embodiment of the present invention.

[0010] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the robot obstacle recognition method according to any embodiment of the present invention.

[0011] The technical solution of this invention acquires echo data from multiple ultrasonic probes during robot movement within a set period, along with the robot's real-time position information. Based on the echo data and the real-time position information, a target acoustic image is determined. When a recognition processing condition is triggered, the target acoustic image is input into a recognition model according to a fixed inference cycle to obtain an initial obstacle recognition result. The initial obstacle recognition results from multiple inference cycles are then filtered to obtain the final target obstacle recognition result. This technical solution, by determining the corresponding acoustic image based on echo data from multiple ultrasonic probes and combining it with a recognition model, achieves obstacle target detection and recognition. It can better identify relevant information about obstacle targets, improves the accuracy of obstacle target recognition, and is less susceptible to interference from environmental visibility.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a robot obstacle recognition method according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of a robot obstacle recognition method according to Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of a robot obstacle recognition device according to Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided according to Embodiment 4 of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0016] It should be noted that the terms "first," "second," "initial," and "target," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] Example 1 Figure 1 This is a flowchart of a robot obstacle recognition method according to Embodiment 1 of the present invention. This embodiment is applicable to the identification of obstacle targets during robot movement. The method can be executed by a robot obstacle recognition device, which can be implemented in hardware and / or software and can be configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes: In this embodiment, multiple ultrasonic probes are deployed on the robot according to set installation conditions. These installation conditions can be preset layout requirements. In this embodiment, the specific layout requirements can be determined based on the required obstacle detection range. Alternatively, specific ultrasonic probe installation requirements can be set according to actual needs. For example, in this embodiment, the blind-spot-free distance can be set to at least 30cm, and the overall field of view (FOV) can be designed to be 120° in both azimuth and pitch. Ultrasonic probes with a uniform FOV of 60° can be used. To ensure effective frontal detection while maximizing coverage of the close-range area, three ultrasonic probes can be installed on the robot. The straight-line distance between any two ultrasonic probes should be less than or equal to 34cm, and the distance between the installed ultrasonic probes and the edge of the robot should be less than 17cm.

[0018] S110. Acquire echo data from multiple ultrasonic probes during the robot's movement within a set period, as well as the robot's real-time position information.

[0019] The set period can be a pre-set data acquisition period or a fixed execution period for constructing the image algorithm. For example, the set period in this embodiment can be 40ms. In this embodiment, each set period processes only one data unit in the probe data buffer. The ultrasonic probe can refer to an ultrasonic detection device installed in the robot according to a preset layout, used to emit and receive ultrasonic signals. For example, the multiple ultrasonic probes in this embodiment can be three ultrasonic probes, and the specific number of ultrasonic probes can be determined according to actual needs. Echo data can refer to the signal received by the probe after the ultrasonic wave emitted by the ultrasonic probe is reflected by an obstacle. In this embodiment, the echo data can include echo distance information and echo intensity, and each echo data carries timestamp information. Real-time position information can refer to the robot's real-time position coordinate information output during movement.

[0020] In this embodiment, during robot operation, the raw echo signals of all ultrasonic probes can be continuously collected at a period of 40ms. Then, the echo signals are analyzed to extract the echo distance and echo intensity corresponding to each probe. Each echo data is then bound with a corresponding timestamp and stored in the probe data buffer. Simultaneously, the real-time spatial coordinates output by the robot positioning module are read and bound with the corresponding timestamp as well.

[0021] Specifically, in this embodiment, an inertial measurement unit (IMU) and a wheeled odometry system can be used to construct a coordinate system with the spatial node where the robot starts as the origin. The robot's movement within this coordinate system (including x-coordinate, y-coordinate, and heading angle a) and its corresponding timestamp are recorded at 10ms intervals and stored in a buffer to obtain real-time position information during robot movement. In this embodiment, ultrasonic probe data can be updated at 20ms intervals and stored in the corresponding buffer. This buffer uses a first-in-first-out (FIFO) queue structure, storing data in data packets composed of the latest data from the current N probes. If probe A's data is not refreshed in the current update, its refresh flag and data will not change.

[0022] S120. Determine the target acoustic image based on echo data and real-time location information.

[0023] The target acoustic image can be a visualized image containing the outline of obstacles, obtained by visualizing the acoustic waves based on echo data and real-time location information. In this embodiment, the target acoustic image can refer to a bird's-eye view generated with the robot's operating environment as the object. By using arcs to represent the ultrasonic detection range and color levels to represent the echo intensity, a visualized image containing the outline of obstacles is formed after accumulating data from multiple probes.

[0024] In this embodiment, the process of determining the target acoustic image can be a process of visualizing the information returned by the ultrasonic echo data in the form of an arc and drawing it on a bird's-eye view. In this process, since the position of the obstacle will continuously reflect the echo, the edge position of the obstacle will gradually accumulate an arc in the image, and as the robot moves, the real-time position information will depict the outline of the obstacle on the image, and finally form a bird's-eye view using color levels to represent the reflection intensity.

[0025] S130. When the recognition processing condition is triggered, the target acoustic image is input into the recognition model according to the fixed inference cycle to obtain the initial obstacle recognition result.

[0026] The recognition processing condition refers to the triggering condition for the recognition model to process the target acoustic image, and it can be pre-set. In this embodiment, the recognition processing condition can be that the cumulative number of updates from multiple probes reaches a set threshold. It is understood that in this embodiment, the image construction process and the model recognition processing are two separate processing threads. Each update of probe data in the image construction thread is recorded. For example, in this embodiment, it can be determined whether the cumulative number of probe updates reaches the set threshold of 100. If the cumulative number of probe updates reaches 100, the model recognition processing thread is triggered to read the current bird's-eye view data and perform model inference, that is, the recognition model can perform the processing operation to recognize the target acoustic image.

[0027] The fixed inference cycle refers to the execution cycle of a single inference by the recognition model. In this embodiment, the fixed inference cycle can be a pre-set model inference cycle; for example, it can be 300ms, but can also be set according to actual needs. Furthermore, in this embodiment, one inference cycle can be defined as one inference frame. The recognition model can be a pre-trained obstacle recognition network model used to recognize obstacle-related information. In this embodiment, the recognition model can input a sound wave image and output corresponding obstacle location and confidence level information. The initial obstacle recognition result can refer to the original detection result of the target sound wave image directly output by the model. In this embodiment, the initial obstacle recognition result can include obstacle location information, size, orientation angle, height attribute, and confidence level information.

[0028] In this embodiment, based on a pre-trained network model, with an execution cycle of 300ms, the target acoustic image obtained can be used to infer the location of the obstacle target. That is, the target acoustic image can be input into the trained recognition model to obtain information such as the location, size, orientation angle, height attribute and confidence level of the obstacle contained in the target acoustic image.

[0029] Specifically, in this embodiment, the number of probe updates accumulated in the image construction thread can be monitored in real time. When the number of probe updates reaches the accumulated threshold, it is determined that the recognition processing conditions are met, the inference process is triggered, and then the target acoustic image of the current frame is input into the trained recognition model according to the preset fixed inference cycle. The model analyzes the image features, detects the obstacle target, and outputs the obstacle position and confidence level data corresponding to the current inference frame, which is the initial obstacle recognition result.

[0030] S140. Filter the initial obstacle recognition results from multiple inference cycles to obtain the target obstacle recognition results.

[0031] The filtering process can be an operation to remove overlaps in the initial obstacle recognition results from multiple inference cycles. In this embodiment, the filtering process can include performing a single inference cycle filtering operation on the initial obstacle recognition results from multiple inference cycles, and managing target associations across multiple inference cycles, followed by an overlap removal operation based on the association results. The target obstacle recognition result can be a stable and effective obstacle detection result output after filtering.

[0032] In this embodiment, the initial obstacle recognition results output during the inference process can be sorted and associated with targets according to their confidence levels. Filtering is then performed based on the sorting results and target association status to obtain the final obstacle recognition results. Specifically, in this embodiment, for obstacles in a single frame (i.e., obstacles in a single inference cycle whose confidence level meets a pre-set confidence threshold), these obstacles are retained; for obstacles whose confidence level does not meet the pre-set confidence threshold, these obstacles are filtered out. For multiple inference cycles (i.e., obstacle targets in multiple frames), data association processing is performed. Then, for a specific obstacle target based on the association results, if the number of frames in which its stable association exists is greater than a set frame threshold, then the obstacle target is saved and output; otherwise, the obstacle target is filtered out.

[0033] The technical solution of this invention acquires echo data from multiple ultrasonic probes during robot movement within a set period, along with the robot's real-time position information. Based on the echo data and real-time position information, a target acoustic image is determined. When a recognition processing condition is triggered, the target acoustic image is input into a recognition model according to a fixed inference cycle to obtain an initial obstacle recognition result. The initial obstacle recognition results from multiple inference cycles are then filtered to obtain the final target obstacle recognition result. This technical solution, by determining the corresponding acoustic image based on echo data from multiple ultrasonic probes and combining it with a recognition model, achieves obstacle target detection and recognition. It can better identify relevant information about obstacle targets, improves obstacle target recognition accuracy, and is less susceptible to interference from environmental visibility.

[0034] Example 2 Figure 2 This is a flowchart of a robot obstacle recognition method according to Embodiment 2 of the present invention. This embodiment is based on the above embodiment and optimized. Specifically, the optimization is as follows: the echo data and real-time position information carry timestamps; the echo data includes echo distance information and echo intensity; determining the target acoustic image based on the echo data and real-time position information includes: determining whether the set period is the first image determination period; if the set period is the first image determination period, then acquiring an initial image of a first size, and determining an initial acoustic image of a second size at the center of the initial image; wherein the second size is smaller than the first size; determining the target position information of the corresponding robot based on the timestamp of the echo data; for multiple ultrasonic probes, determining the target acoustic image based on the target position information, echo distance information, and echo intensity. Figure 2 As shown, the method includes: S210. Acquire echo data from multiple ultrasonic probes during the robot's movement within a set period, as well as the robot's real-time position information.

[0035] In this embodiment, the echo data and real-time location information carry timestamps; the echo data includes echo distance information and echo intensity.

[0036] The timestamp of the echo data refers to the time information corresponding to the receipt of the echo data. The timestamp of the real-time location information refers to the time information corresponding to the acquisition of the real-time location information. In this embodiment, the timestamp information can be used to match ultrasonic detection data and robot coordinates at the same time. The echo distance information refers to the distance value from the ultrasonic probe to the obstacle detected by the ultrasonic probe. The echo intensity refers to the amplitude of the echo signal after the ultrasonic wave is reflected by the obstacle, which can be used to characterize the strength of the reflection and corresponds to the image color gamut value.

[0037] S220. Determine whether the set period is the first image determination period.

[0038] The image determination cycle can refer to the cycle of determining the target acoustic image in a single round, i.e., the execution cycle of a single round of image determination. In this embodiment, the first image determination cycle can refer to the first image determination cycle after the robot starts. In this embodiment, the image construction process can be started every 40ms, and the current cycle number is marked during runtime; the cycle mark is read to determine whether the current cycle is the first image determination cycle after the robot is powered on.

[0039] S230. If the set period is the first image determination period, then obtain the initial image of the first size, and determine the initial sound wave image of the second size at the center of the initial image.

[0040] The second dimension is smaller than the first dimension. The first dimension can be the pre-defined size information of the global initialization image. For example, in this embodiment, the first dimension can be 40m × 40m, but it can also be set according to actual needs. Initialization image: A global bird's-eye view base map created when the robot starts, which can serve as the basic canvas for the entire detection area. The initial acoustic image can refer to a local image nested at the center of the global image, which is the core area for arc drawing and contour accumulation. In this embodiment, the second dimension can be the pre-defined size information of the initial acoustic image. For example, in this embodiment, the second dimension can be 20m × 20m, but it can also be set according to actual needs.

[0041] Understandably, in this embodiment, the image determination process is performed with a period of 40ms, and one data unit in the probe data buffer is processed in each period.

[0042] In this embodiment, it is determined whether the current cycle is the first image determination cycle after the robot starts. When it is determined to be the first image determination cycle, a blank global initialization image of 40m×40m is first created; then, a rectangular area of ​​20m×20m is delineated as the initial acoustic wave image based on the center point of the global image.

[0043] In this embodiment, optionally, it further includes: if the set period is not the first image determination period, then determining whether the real-time location information exceeds the range boundary of the first size; if the real-time location information exceeds the range boundary of the first size, then updating the initial image of the first size based on the real-time location information to obtain the updated image.

[0044] The range boundary can refer to the four boundaries of the initial image of the first size, which can be used to determine whether the robot has left the current map coverage area. In this embodiment, updating the initial image can be done by regenerating the global map after the robot crosses the boundary, in order to retain the dynamic update operation of the valid historical area. The updated image can refer to the new first-size global image obtained after completing the recentering of the position and the overlay of historical areas.

[0045] In this embodiment, when it is determined that the current cycle is not the first image determination cycle after the robot starts, the robot's current real-time spatial coordinates are retrieved; the coordinate values ​​are compared with the preset coordinates of the four boundaries of the initial image; it is determined whether the robot exceeds the map boundary of the first size in the X and Y axes. If it is confirmed that the robot's position has exceeded the original map boundary, the map update process is triggered. An image mask of the same size is created with the robot's current real-time position as the new center, and a valid historical region of 20m×40m behind the robot in the original initial image is extracted and superimposed on the new image mask; the region superposition and image rendering are completed, a new global image is generated to replace the original initial image; and the map coordinate reference and boundary parameters are updated. The image mask can be a new image canvas generated with the robot's current position as the center, serving as the base of the updated global map.

[0046] Furthermore, in this embodiment, if the real-time location information does not exceed the range boundary of the first size, the original initial image will continue to be used.

[0047] In this embodiment, the global image is automatically updated when the robot moves beyond the original map boundary, ensuring that the robot is always within the effective range of the map and avoiding interruption of image determination and loss of target due to going out of bounds; and when updating the map, the image of the area behind the robot is superimposed, completely preserving the outline of the detected obstacles, without the need for repeated scanning and mapping, thus improving the efficiency of acoustic image determination.

[0048] S240. Determine the target position information of the corresponding robot based on the timestamp of the echo data.

[0049] The target location information can refer to the robot's spatial coordinates that match the current echo data in time sequence.

[0050] In this embodiment, multiple echo data collected in the current cycle and their corresponding timestamps can be retrieved, and the real-time position sequence of the robot with timestamps can be read. Based on the timestamp of a single echo data, the coordinate data with the smallest time difference is matched in the position information. The matched robot coordinates are used as the target position information and bound to the corresponding echo data.

[0051] Understandably, in this embodiment, for the data of a probe, it is determined whether it has been updated. If it has been updated, the spatial coordinates with the closest timestamp in the buffer are matched.

[0052] S250: For multiple ultrasonic probes, determine the target acoustic image based on target location information, echo distance information, and echo intensity.

[0053] The target acoustic image can refer to the final bird's-eye view formed by overlaying all probe detection arcs onto the initial acoustic image. In this implementation, the target acoustic image includes obstacle outlines and echo reflection intensity information.

[0054] In this embodiment, the valid data of all ultrasonic probes in the current cycle can be traversed. Then, with the matched target location information as the center and the echo distance as the radius, the central angle of the arc is calculated according to the preset segmented formula. The echo intensity is mapped to the color gamut value, and the corresponding arc is drawn in the initial acoustic image in combination with the central angle. All probes complete the arc drawing in sequence, and the graphics are continuously superimposed and accumulated to gradually outline the outline of the obstacle. All the drawn content is integrated to generate the final target acoustic image, and the number of probe data updates is counted simultaneously.

[0055] In this embodiment, optionally, determining the target acoustic image based on the target location information, echo distance information, and echo intensity includes: determining the corresponding central angle based on the target location information and echo distance information; determining the corresponding arc information based on the central angle and echo intensity; and generating arc information sequentially based on echo data from multiple probes within a set period and superimposing it onto the initial acoustic image to obtain the target acoustic image.

[0056] The central angle can refer to the angle within the ultrasonic detection sector, representing the effective detection range of a single probe, and its value dynamically changes with the echo distance. The arc information can be an ultrasonic detection arc containing attributes such as the center, radius, central angle, and color gamut value.

[0057] In this embodiment, the target location information is used as the center and the echo distance is used as the basic parameter. A preset segmented calculation formula is called, and the interval to which the current echo distance belongs is determined in combination with the current echo distance to calculate the central angle corresponding to the current probe.

[0058] In this embodiment, the determined central angle, echo distance (radius), and target position (center) can be used as geometric parameters; the echo intensity is converted into image color gamut parameters to characterize the strength of echo reflection; all parameters are integrated to generate complete single-probe arc drawing data, i.e., arc information. In this embodiment, the echo data of all ultrasonic probes within the current 40ms cycle can be traversed, and the first two steps can be repeated to generate corresponding arc information one by one; each arc is drawn sequentially and superimposed on the initial acoustic image; through the continuous accumulation of multiple arcs, relying on the characteristic of the obstacle continuously reflecting echoes, the shape of the obstacle is gradually outlined; after all probe data is processed, the corresponding target acoustic image is obtained.

[0059] Furthermore, in this embodiment, the specific method for determining the corresponding central angle based on the target position information and echo distance information can be to use the robot's spatial position coordinates as the center and the distance to the obstacle returned by the ultrasonic probe as the radius, and calculate the central angle using a preset segmented calculation formula. The returned echo intensity (LSB) is used as the color gamut value to draw an arc in the bird's-eye view. The preset segmentation calculation formula can be: ; Where dis is the distance to the obstacle returned by the probe; arccos is the inverse cosine function; pi is pi; and round is the rounding function used for rounding.

[0060] Understandably, in this embodiment, every 40ms set period, the arcs in the bird's-eye view will be updated. Whether the specific obstacle content has been updated depends on the specific data detected by the probes. The steps of determining the target location information and central angle are repeated for the N probe data in this cache unit, and the number of probe updates is accumulated. When it is determined whether the accumulated number of probe updates has reached the set threshold of 100, if it reaches 100, the inference thread is triggered to read the current bird's-eye view data for inference.

[0061] In addition, in this embodiment, it is also possible to determine whether the execution time of the current image determination process is greater than 40ms. If the execution time is less than 40ms, the process enters a waiting phase so that the entire execution cycle reaches 40ms, thereby ensuring the stability of the algorithm operation.

[0062] In this embodiment, the central angle can be dynamically calculated based on the detection distance, which conforms to the actual detection characteristics of the ultrasonic probe and improves the authenticity and reliability of the image. Moreover, by integrating the three types of information—spatial position, detection distance, and echo intensity—into the same image, it combines multiple information dimensions and improves the input quality of the recognition model.

[0063] In this embodiment, each set cycle can convert the currently updated ultrasonic echo into a visualized arc and draw it on the bird's-eye view. Furthermore, the position detection operation at the beginning of the algorithm can ensure that the bird's-eye view can be continuously updated to follow the robot's movement.

[0064] S260. When the recognition processing condition is triggered, the target acoustic image is input into the recognition model according to the fixed inference cycle to obtain the initial obstacle recognition result.

[0065] In this embodiment, optionally, when the recognition processing condition is triggered, the target acoustic image is input into the recognition model according to a fixed inference cycle to obtain the initial obstacle recognition result, including: when the recognition processing condition is triggered, the size of the target acoustic image is adjusted to obtain the input acoustic image; the input acoustic image is input into the recognition model to obtain the initial obstacle recognition result.

[0066] The resizing process involves adjusting the size of the generated acoustic image to the fixed input size required by the recognition model. The input acoustic image can be a standardized acoustic image that meets the model's input format and size requirements.

[0067] In this embodiment, the number of probe updates or the amount of image data during the image determination process can be monitored in real time. When a preset threshold is reached, the recognition processing condition is triggered. The target acoustic image that has been generated is read, and according to the preset model input size requirements, scaling and size normalization are performed on the target acoustic image to obtain an input acoustic image that can be directly input into the model. The input acoustic image with the adjusted size is sent into the recognition model, and the model performs forward reasoning such as feature extraction and target detection on the image to obtain the initial obstacle recognition result containing the obstacle position and confidence level.

[0068] In this embodiment, the inference process is triggered only if the current inference thread has been running for more than 300ms and is in an idle state. After entering the triggering process, the 40m×40m bird's-eye view is first updated based on the latest robot coordinates at the time of inference entry, ensuring that the bird's-eye view is the latest acoustic image at the current time and that the robot's position has not exceeded the boundary of the bird's-eye view. Then, using the current position of the robot in the bird's-eye view as the origin, a 20m×20m bird's-eye view is cropped from the 40m×40m view, and the image size is scaled and quantized from 16 bits to 8 bits. The bird's-eye view is then input into the network model for inference to obtain the initial obstacle recognition result.

[0069] In this embodiment, the obstacle location information of the initial obstacle recognition result can be represented in the form of a target rectangular detection box. Furthermore, after obtaining the initial obstacle recognition result, this embodiment can also use a non-maximum suppression algorithm and a set confidence threshold to filter the output obstacles, clustering multiple overlapping target detection boxes into a single target and outputting obstacles with higher confidence.

[0070] In this embodiment, the input sound wave image can be standardized in image format, and the size can be adjusted to ensure that the image matches the model input, avoiding inference failure due to size incompatibility, thus ensuring the stable operation of the recognition process.

[0071] S270. Filter the initial obstacle recognition results from multiple inference cycles to obtain the target obstacle recognition results.

[0072] In this embodiment, optionally, the initial obstacle recognition result includes obstacle location information and confidence information; one inference cycle is one inference frame; filtering the initial obstacle recognition results of multiple inference cycles to obtain the target obstacle recognition result includes: for the initial obstacle recognition result obtained in a single inference frame, performing a filtering operation on the initial obstacle recognition result based on confidence information and preset threshold conditions to obtain a first obstacle recognition result; associating the first obstacle recognition result corresponding to the current inference frame with the second obstacle recognition result of the previous inference frame to obtain the target obstacle recognition result.

[0073] The inference cycle can be the execution cycle of a single inference by the recognition model, generating one frame of data after each inference. The filtering operation can be based on confidence judgment rules to remove interfering targets whose built-in confidence level in a single frame is insufficient. The preset threshold condition can be a pre-set confidence threshold condition, which can be used as the judgment standard for single-frame target filtering. In this embodiment, the preset threshold condition can be a confidence threshold curve set based on different distances. The first obstacle recognition result can be the valid obstacle target data obtained after the obstacle target recognition result of the current inference frame has been filtered by confidence. The second obstacle recognition result can be the valid obstacle target data cached after the filtering of the previous inference frame. The association processing can be a processing operation that compares obstacle targets in previous and subsequent frames, matches identical obstacle targets, and thus associates identical obstacle targets.

[0074] In this embodiment, all initial obstacle recognition results of the current inference frame can be extracted, and the confidence level and detection distance of each obstacle target can be obtained one by one. Based on the target detection distance, a preset distance and confidence threshold curve is queried to obtain the confidence judgment threshold corresponding to that distance. The actual confidence level of the obstacle target is compared with the confidence judgment threshold. If the confidence level of the obstacle target is greater than or equal to the corresponding confidence judgment threshold, the obstacle target is retained; otherwise, it is determined as a false detection and the obstacle target is removed. After completing the filtering of all obstacle targets in the current frame, the retained data is the first obstacle recognition result. Then, the cached second obstacle recognition result from the previous frame is retrieved simultaneously. By traversing all obstacle detection boxes in the two sets of data, the intersection-union ratio (IUU) is calculated pairwise to determine the degree of region overlap, and then obstacle association processing is performed to obtain the associated obstacle recognition result. Finally, the associated obstacle recognition result is filtered again according to the obstacle target management rules to obtain the final target obstacle recognition result.

[0075] In this embodiment, the obstacle targets (ODs) output by the inference results can also be filtered based on confidence threshold curves set at different distances; for obstacle targets that intersect with the robot's position, the target attributes, confidence, and intersection-union ratio (IOU) with the robot's position are combined for filtering; finally, for the remaining obstacle targets after filtering, the coordinate system is transformed so that they enter the coordinate system centered on the robot.

[0076] In this embodiment, the setting allows for the rapid filtering of false targets caused by model misdetection and environmental interference through confidence screening. By using obstacle targets to correlate each inference frame with each other, real obstacles and instantaneous noise can be distinguished, avoiding the impact of occasional false detections in a single frame on the final obstacle recognition result, and greatly improving the accuracy and stability of obstacle recognition.

[0077] In this embodiment, optionally, obstacle location information is represented by rectangular detection box coordinates; the first obstacle recognition result corresponding to the current inference frame and the second obstacle recognition result of the previous inference frame are associated to obtain the target obstacle recognition result, including: determining the intersection-union ratio (IUR) of the first rectangular detection box coordinates corresponding to the current inference frame and the second rectangular detection box coordinates of the previous inference frame; if the IUR exceeds a preset IUR threshold, the first obstacle recognition result corresponding to the current inference frame and the second obstacle recognition result of the previous inference frame are associated to obtain the associated obstacle recognition result; determining the number of consecutive associated frames of the associated obstacle recognition result; if the number of consecutive associated frames is greater than a preset frame number threshold, the associated obstacle recognition result is determined as the target obstacle recognition result.

[0078] The first rectangular detection box coordinates can refer to the rectangular detection box coordinates used to represent the obstacle's position information in the first obstacle recognition result of the current inference frame. The second rectangular detection box coordinates can refer to the rectangular detection box coordinates used to represent the obstacle's position information in the second obstacle recognition result of the previous inference frame. The Intersection over Union (IOU) ratio can be an overlap index calculated based on the coordinates of the two rectangular detection boxes, used to determine whether the target in the preceding and following frames is the same obstacle. The preset IOU threshold can be a pre-set IOU determination threshold used to determine whether the target in the preceding and following frames belongs to the same obstacle.

[0079] In this embodiment, data association can be an operation that matches and binds the identification results of the same obstacle in two consecutive frames to form a continuous target trajectory. The associated obstacle identification results can be continuous obstacle target identification information formed by matching and binding between any two inference frames. The number of consecutive associated frames can refer to the number of frames in which the same obstacle is successfully associated and continuously detected in multiple consecutive inference frames. The preset frame number threshold can be a pre-set threshold for the number of consecutive associated frames, which can be set according to actual needs. For example, the preset frame number threshold in this embodiment is 15 frames; only when this threshold is reached can it be determined as a real and valid obstacle. In this embodiment, the target obstacle identification result can be the stable and reliable valid obstacle identification information finally output after single-frame filtering, multi-frame association, and continuous frame number verification.

[0080] In this embodiment, the coordinates of a first rectangular detection box can be extracted from the first obstacle recognition result of the current inference frame; the coordinates of a second rectangular detection box can be extracted from the second obstacle recognition result of the previous inference frame; the intersection area and union area are calculated based on the coordinates of the two sets of rectangular detection boxes; the corresponding intersection-union ratio (IUR) is obtained based on the area ratio; the calculated IUR is compared with a preset IUR threshold; if the IUR is greater than the IUR threshold, the obstacle in the current frame and the obstacle in the previous frame are determined to be the same obstacle; data association is performed on the two sets of obstacle recognition results to bind them to the same continuous target, generating the associated obstacle recognition result, i.e., the associated obstacle recognition result. If the IUR is less than the IUR threshold, it means that they are not the same obstacle, and data association is not performed.

[0081] In this embodiment, for the obstacle identification results that have been associated, the continuous association count in the previous frame is queried; the count is automatically incremented by 1 based on the original continuous association frame count to update to the latest continuous association frame count; the latest continuous association frame count corresponding to the associated obstacle is saved and recorded. The continuous association frame count of the current associated obstacle is compared with a preset frame count threshold; if the continuous association frame count is greater than the preset frame count threshold, the obstacle is determined to be a real and valid target; the associated obstacle identification result is officially determined as the final target obstacle identification result.

[0082] For example, in this embodiment, for the obstacle target OD in the current frame, the OD data saved in the previous frame can be traversed, and the association calculation can be performed based on the IOU of the two frames of OD. ODs that meet the IOU threshold are formed into association pairs. Then, for the same OD data, if the OD data can be continuously associated for more than 15 frames, the OD is called a stable OD and is output externally.

[0083] In this embodiment, each time a stable OD is associated, a weighted filter is used to update the OD's center position, size, orientation angle, elevation attributes, and confidence level. When a stable OD fails to associate, the OD is extrapolated based on the robot's motion coordinates corresponding to the current inference frame. If an OD fails to associate for four consecutive inference frames, it is determined that the OD has disappeared, and it will no longer be output.

[0084] In this embodiment, the robot's movement route planning can be optimized based on the output target obstacle recognition results.

[0085] In this embodiment, the intersection-union ratio (IUGR) can be calculated based on the coordinates of the rectangular detection box to accurately match the same obstacle in consecutive frames and construct a continuous target trajectory. By filtering through IUGR and consecutive associated frame counts, instantaneous false detections and interfering targets are effectively filtered out, ensuring that the output obstacle recognition results are stable and reliable.

[0086] In this embodiment, a coordinate system is established based on the robot's movement position information, and a map is constructed based on the echo of the ultrasonic sensor to obtain a sound wave image. Then, the sound wave image is processed by the corresponding network model to accurately determine the position of obstacles in the current environment. This enables precise detection of obstacle targets within a radius of 20m centered on the robot, unaffected by environmental visibility (such as rain, snow, and dust). It also covers the blind spots of the lidar at close range, enhancing the overall robustness of obstacle detection.

[0087] Example 3 Figure 3 This is a schematic diagram of a robot obstacle recognition device according to Embodiment 3 of the present invention. The robot is equipped with multiple ultrasonic probes deployed according to pre-defined installation conditions. Figure 3 As shown, the device includes: The data acquisition module 310 is used to acquire echo data from multiple ultrasonic probes during the robot's movement within a set period, as well as the robot's real-time position information. Image determination module 320 is used to determine the target acoustic image based on echo data and real-time location information; The obstacle recognition module 330 is used to input the target acoustic image into the recognition model according to a fixed inference cycle when the recognition processing condition is triggered, so as to obtain the initial obstacle recognition result. The filtering module 340 is used to filter the initial obstacle recognition results of multiple inference cycles to obtain the target obstacle recognition results.

[0088] Optionally, the echo data and real-time location information carry timestamps; the echo data includes echo distance information and echo intensity; the image determination module 320 includes: The judgment unit is used to determine whether the set period is the first image determination period; An initial image determination unit is used to acquire an initial image of a first size and determine an initial acoustic image of a second size at the center of the initial image if the set period is the first image determination period; wherein the second size is smaller than the first size; The position determination unit is used to determine the target position information of the corresponding robot based on the timestamp of the echo data. The target image determination unit is used to determine the target acoustic image based on the target location information, echo distance information, and echo intensity for multiple ultrasonic probes.

[0089] Optional, also includes: The boundary judgment unit is used to determine whether the real-time position information exceeds the range boundary of the first size if the set period is not the first image determination period. The update unit is used to update the initial image of the first size based on the real-time location information if the real-time location information exceeds the range boundary of the first size, so as to obtain the updated image.

[0090] Optionally, the target image determination unit is specifically used for: Determine the corresponding central angle based on the target location information and echo distance information; The corresponding arc information is determined based on the central angle and echo intensity; Based on the echo data from multiple probes within a set period, arc information is generated sequentially and superimposed onto the initial acoustic image to obtain the target acoustic image.

[0091] Optional, obstacle recognition module 330, specifically used for: When the recognition processing condition is triggered, the size of the target acoustic wave image is adjusted to obtain the input acoustic wave image; Input the sound wave image into the recognition model to obtain the initial obstacle recognition result.

[0092] Optionally, the initial obstacle recognition result includes obstacle location information and confidence information; one inference cycle is one inference frame; the filtering processing module 340 includes: The filtering unit is used to filter the initial obstacle recognition results obtained from a single inference frame based on confidence information and preset threshold conditions to obtain the first obstacle recognition result. The association unit is used to associate the first obstacle recognition result corresponding to the current inference frame with the second obstacle recognition result of the previous inference frame to obtain the target obstacle recognition result.

[0093] Optionally, obstacle location information can be represented using the coordinates of a rectangular detection box; Associated units, specifically used for: Determine the intersection-union ratio (IUU) of the coordinates of the first rectangular detection box corresponding to the current inference frame and the coordinates of the second rectangular detection box of the previous inference frame; If the cross-union ratio exceeds the preset cross-union ratio threshold, the first obstacle recognition result corresponding to the current inference frame and the second obstacle recognition result of the previous inference frame are associated to obtain the associated obstacle recognition result. Determine the number of consecutive associated frames for the associated obstacle recognition results; If the number of consecutive associated frames exceeds a preset frame threshold, the associated obstacle recognition result will be determined as the target obstacle recognition result.

[0094] The robot obstacle recognition device provided in this embodiment of the invention can execute the robot obstacle recognition method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0095] Example 4 Figure 4 This is a schematic diagram of an electronic device according to Embodiment 4 of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0096] like Figure 4 As shown, the electronic device 410 includes at least one processor 411 and a memory, such as a read-only memory (ROM) 412 or a random access memory (RAM) 413, communicatively connected to the at least one processor 411. The memory stores computer programs executable by the at least one processor. The processor 411 can perform various appropriate actions and processes based on the computer program stored in the ROM 412 or loaded from storage unit 418 into the RAM 413. The RAM 413 can also store various programs and data required for the operation of the electronic device 410. The processor 411, ROM 412, and RAM 413 are interconnected via a bus 414. An input / output (I / O) interface 415 is also connected to the bus 414.

[0097] Multiple components in electronic device 410 are connected to I / O interface 415, including: input unit 416, such as keyboard, mouse, etc.; output unit 417, such as various types of displays, speakers, etc.; storage unit 418, such as disk, optical disk, etc.; and communication unit 419, such as network card, modem, wireless transceiver, etc. Communication unit 419 allows electronic device 410 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0098] Processor 411 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 411 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 411 performs the various methods and processes described above, such as robot obstacle recognition methods.

[0099] In some embodiments, the robot obstacle recognition method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 418. In some embodiments, part or all of the computer program may be loaded into and / or mounted on electronic device 410 via ROM 412 and / or communication unit 419. When the computer program is loaded into RAM 413 and executed by processor 411, one or more steps of the robot obstacle recognition method described above may be performed. Alternatively, in other embodiments, processor 411 may be configured to perform the robot obstacle recognition method by any other suitable means (e.g., by means of firmware).

[0100] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0101] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0102] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0103] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0104] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0105] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0106] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0107] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A robot obstacle recognition method, characterized in that, The robot is equipped with multiple ultrasonic probes deployed according to pre-defined installation conditions; the method includes: Acquire echo data from multiple ultrasonic probes during robot movement within a set period, as well as the robot's real-time position information; The target acoustic image is determined based on the echo data and the real-time location information; When the recognition processing condition is triggered, the target acoustic image is input into the recognition model according to a fixed inference cycle to obtain the initial obstacle recognition result. The initial obstacle recognition results from multiple inference cycles are filtered to obtain the target obstacle recognition results.

2. The method according to claim 1, characterized in that, The echo data and the real-time location information carry timestamps; the echo data includes echo distance information and echo intensity. Determining the target acoustic image based on the echo data and the real-time location information includes: Determine whether the set period is the first image determination period; If the set period is the first image determination period, then an initial image of the first size is obtained, and an initial acoustic image of the second size is determined at the center of the initial image; wherein, the second size is smaller than the first size; The target position information of the corresponding robot is determined based on the timestamp of the echo data; For the plurality of ultrasonic probes, a target acoustic image is determined based on the target location information, the echo distance information, and the echo intensity.

3. The method according to claim 2, characterized in that, Also includes: If the set period is not the first image determination period, then determine whether the real-time location information exceeds the range boundary of the first size; If the real-time location information exceeds the boundary of the first size, the initial image of the first size is updated based on the real-time location information to obtain the updated image.

4. The method according to claim 2, characterized in that, Determining the target acoustic image based on the target location information, the echo distance information, and the echo intensity includes: The corresponding central angle is determined based on the target location information and the echo distance information; The corresponding arc information is determined based on the central angle and the echo intensity; Based on the echo data from multiple probes within the set period, arc information is generated sequentially and superimposed onto the initial acoustic image to obtain the target acoustic image.

5. The method according to claim 1, characterized in that, When the recognition processing condition is triggered, the target acoustic image is input into the recognition model according to a fixed inference cycle to obtain the initial obstacle recognition result, including: When the recognition processing condition is triggered, the size of the target acoustic image is adjusted to obtain the input acoustic image; The input acoustic image is input into the recognition model to obtain the initial obstacle recognition result.

6. The method according to claim 1, characterized in that, The initial obstacle identification result includes obstacle location information and confidence information; one inference cycle is one inference frame; The initial obstacle recognition results from multiple inference cycles are filtered to obtain the target obstacle recognition results, including: For the initial obstacle recognition result obtained from a single inference frame, the initial obstacle recognition result is filtered based on the confidence information and a preset threshold condition to obtain a first obstacle recognition result; The first obstacle recognition result corresponding to the current inference frame and the second obstacle recognition result of the previous inference frame are correlated to obtain the target obstacle recognition result.

7. The method according to claim 6, characterized in that, The obstacle location information is represented by the coordinates of a rectangular detection box; The first obstacle recognition result corresponding to the current inference frame and the second obstacle recognition result of the previous inference frame are correlated to obtain the target obstacle recognition result, including: Determine the intersection-union ratio (IUU) of the coordinates of the first rectangular detection box corresponding to the current inference frame and the coordinates of the second rectangular detection box of the previous inference frame; If the cross-union ratio exceeds a preset cross-union ratio threshold, the first obstacle recognition result corresponding to the current inference frame and the second obstacle recognition result of the previous inference frame are associated to obtain an associated obstacle recognition result. Determine the number of consecutive associated frames of the associated obstacle recognition results; If the number of consecutive associated frames is greater than a preset frame threshold, then the associated obstacle recognition result is determined as the target obstacle recognition result.

8. A robot obstacle recognition device, characterized in that, The robot is equipped with multiple ultrasonic probes deployed according to pre-defined installation conditions; the device includes: The data acquisition module is used to acquire echo data from multiple ultrasonic probes during the robot's movement within a set period, as well as the robot's real-time position information. The image determination module is used to determine the target acoustic image based on the echo data and the real-time location information; The obstacle recognition module is used to input the target acoustic image into the recognition model according to a fixed inference cycle when the recognition processing condition is triggered, so as to obtain the initial obstacle recognition result. The filtering module is used to filter the initial obstacle recognition results from multiple inference cycles to obtain the target obstacle recognition result.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the robot obstacle recognition method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the robot obstacle recognition method according to any one of claims 1-7.

Images