Method for Detecting Underwater Obstacles

Abstract

An object of the present invention is to provide a method for detecting an underwater obstacle to solve the problem that the calculation in the warning of the conventional underwater obstacle becomes complicated and further misjudgment occurs. 【Solution means】A zoning step of extending from a plurality of borders of the mobile carrier to the outside of the mobile carrier to form a recognition area, wherein the plurality of recognition areas are continuous and surround the periphery of the mobile carrier; an obstacle recognition step of taking an around-view image centered on the mobile carrier and recognizing at least one obstacle distributed in the plurality of recognition areas in the around-view image; and a distance calculation step of calculating the distance between each obstacle in each recognition area and the border corresponding to the recognition area, that is, the relative distance between the obstacle detected in each recognition area and the mobile carrier, and continuously calculating the plurality of relative distances detected in the plurality of recognition areas during the process of the mobile carrier moving.

Description

Technical Field

[0001] The present invention relates to a collision warning technology in water transportation, and particularly to a method for detecting water obstacles that accurately calculates the relative distance and warns of collisions at multiple points.

Background Art

[0002] Collision prevention detection technology is often applied to carriers such as vehicles, drones, and robots, and is used to warn of obstacles on the moving path or automatically avoid obstacles. The carrier detects the distance between the main body and surrounding objects by radar, predicts objects that may collide in advance, or uses a camera to capture an image around the carrier, and recognizes information such as the boundaries, orientations, and distances of different obstacles by image recognition technology, and transmits to the driver or navigation system to control the speed and moving direction of the carrier, thereby maintaining a safe distance and ensuring a reaction time to achieve the effect of reducing collision accidents.

[0003] However, large ships sailing on water usually cannot quickly change direction, and furthermore, due to the huge volume of the hull, there will be a difference in the judgment of the distance between the bow and stern and the obstacle. Therefore, even if the front half of the hull can avoid the obstacle safely, the rear half may collide due to judgment errors. Therefore, the conventional method for detecting water obstacles needs to improve the accuracy of distance measurement from obstacles and install a plurality of sensors on the hull to collect obstacle information in each orientation facing from different positions of the hull. However, since a large amount of detected obstacle information must be matched and adapted based on detection spots at different positions, the calculation of distance measurement becomes complicated and calculation errors are likely to occur. In view of the above, the conventional method for detecting water obstacles needed to be improved.

Prior Art Documents

Patent Documents

[0004] [Patent Document 1] Chinese Patent Publication No. 109919026 [Summary of the Invention] [Problems to be Solved by the Invention]

[0005] In order to solve the above problems, an object of the present invention is to provide a method for detecting an underwater obstacle that can accurately calculate the distance from an obstacle.

[0006] Furthermore, an object of the present invention is to provide a method for detecting an underwater obstacle that can warn of collisions at multiple points.

[0007] Furthermore, an object of the present invention is to provide a method for detecting an underwater obstacle that can reduce the complexity of calculations. [Means for Solving the Problems]

[0008] The directional terms or their approximate terms described throughout the specification of the present invention, for example, "front", "rear", "left", "right", "top", "bottom", "inner", "outer", "side surface", etc., are based on the directions on the attached drawings, and each directional term or its approximate terms are only for assisting the description and understanding of each embodiment of the present invention, and do not limit the present invention.

[0009] Throughout the specification of the present invention, the numeral words such as "one" or "a" used for parts or components are used for convenience and give the normal meaning to the scope included in the present invention. In the present invention, it should be interpreted as one or at least one, and the concept of one also includes the case of multiple unless specifically indicating another meaning.

[0010] The method for detecting water obstacles according to the present invention divides the contour of the main body of the mobile carrier in the two-dimensional horizontal plane into a plurality of continuous borders, extends from each border to the outside of the mobile carrier to form a recognition area, and the plurality of recognition areas are continuous and surround the mobile carrier, including an area division step; centering on the mobile carrier, taking an around-view image, and recognizing at least one obstacle distributed in the plurality of recognition areas in the around-view image, including an obstacle recognition step; calculating the distance between each obstacle in each recognition area and the border corresponding to the recognition area, that is, the relative distance between the obstacle detected in each recognition area and the mobile carrier, and continuously calculating the plurality of relative distances detected in the plurality of recognition areas during the process of the mobile carrier moving, including a distance calculation step.

[0011] Thereby, the method for detecting water obstacles of the present invention divides the surrounding detection range into areas, classifies obstacles by using the around-view image, detects a plurality of obstacles respectively by using a plurality of recognition areas, or reduces the complexity of calculation in distance measurement by issuing a warning for the same obstacle. Furthermore, the accuracy of detection and distance measurement can be improved by calculating the distance to the obstacle based on different parts and angles of the main body of the mobile carrier, and the effect of providing collision warnings at multiple points can be achieved.

[0012] Also, it is characterized in that a semantic segmentation model is used to classify all pixels in the around-view image and classify the pixels belonging to a plurality of obstacles. Thereby, the semantic segmentation model can classify all pixels in the around-view image, determine whether each pixel is an obstacle or not, and does not require further image processing, achieving the effect of improving the accuracy of detection.

[0013] Also, each of the borders has a plurality of sampling points at equal intervals, and each relative distance is the shortest among all the distances sampled between the plurality of pixels of each obstacle and the plurality of sampling points. This enables determination of the shortest distance from the border of the main body to the edge of the obstacle, achieving the effect of accurately measuring the distance to the obstacle.

[0014] Also, a plurality of relative distances are calculated at regular intervals during the sampling period, and a warning threshold is set as the warning lower limit for the plurality of relative distances. This enables determination of the warning range by adopting a plurality of detection data and comparing with the warning threshold, reducing false judgment in detection, and providing a collision warning.

[0015] Also, the sampling period is 0.3 to 0.5 seconds, the interval time is 0.03 to 0.05 seconds, and the warning threshold is 1 meter. This enables collection and detection of distances in a certain time period, and confirmation of the warning range, ensuring in advance the reaction distance and time for a collision.

[0016] Also, when the data that is less than or equal to the warning threshold among the plurality of relative distances within the sampling period is above the risk ratio, a collision warning is issued. This enables determination that there is a risk of collision when continuously collecting the detection distance within the danger range, avoiding false alarms of the collision warning.

[0017] Also, the risk ratio is 60% to 80%. This enables adjustment of the warning sensitivity by setting a higher risk ratio to lower the warning sensitivity, and setting a lower risk ratio to increase the warning sensitivity, and can be adjusted to an appropriate warning state based on the surrounding environment and movement status of the mobile carrier, adjusting the warning state and reducing false judgment.

[0018] Also, when the average value of the plurality of relative distances within the sampling period is less than or equal to the warning threshold, a collision warning is issued. As a result, the state of the distance change of the obstacle during the sampling period can be used to judge the movement history of the moving carrier, and it has the effect of assisting the warning of the collision risk.

[0019] Further, when the change rate of a plurality of relative distances with respect to time is less than -1 meter / second or equivalent to -1 meter / second, a collision warning is issued. As a result, the above change rate represents the speed at which the moving carrier approaches the obstacle, can be used for predicting a collision, and can secure in advance a reaction time for avoiding the obstacle by warning in advance.

Brief Description of the Drawings

[0020]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Modes for Carrying Out the Invention

[0021] In order to more easily understand the above object, other objects and advantages of the present invention, embodiments of the present invention are given as follows and will be described in detail with reference to the drawings. Also, in different drawings, those denoted by the same reference numerals are regarded as the same and their description is omitted.

[0022] As shown in FIGS. 1 and 2, an embodiment of the method for detecting an underwater obstacle according to the present invention includes an area division step S1, an obstacle recognition step S2, and a distance calculation step S3, with a moving carrier V navigating on the water as a reference point, and a plurality of obstacles B around this moving carrier V are detected.

[0023] The area division step S1 divides the contour of the main body of the mobile carrier V in the two-dimensional horizontal plane into a plurality of borders R. The plurality of borders R have their starting points and ending points connected. Each border R extends to the outside of the mobile carrier V to form a recognition area Z. The recognition areas Z are continuous and surround the mobile carrier V. In this embodiment, the rectangular contour of a large ship is described. As shown in FIG. 2, the plurality of recognition areas Z can include the left bow, the right bow, the left side of the hull, the right side of the hull, the left stern, and the right stern, but the present invention is not limited thereto.

[0024] As shown in FIGS. 3 to 5, the area division step S1 can select a plurality of different borders R and adjust the ranges included in the plurality of recognition areas Z based on the navigation status of the mobile carrier V and the environment in which it is located. As shown in FIG. 3, during the sea navigation of a ship, since the probability of encountering an obstacle is low and the speed at which the obstacle approaches from the forward direction is fast, one recognition area Z1 is directed only to the bow area to intensively detect the sea surface ahead, and another recognition area Z2 can include both sides and the rear of the hull. Also, as shown in FIG. 4, when the ship is docking, it is necessary to adjust the position and angle of the hull by changing direction or reversing, and to get as close to the shore as possible. Therefore, the recognition area Z1 can be kept directed towards the bow, the right side and the right rear of the hull can be assigned to the recognition area Z2, and the left side and the left rear of the hull can be assigned to the recognition area Z3. Thereby, during the process of the ship docking at the shore, the distance between the ship and the shore can be detected. As shown in FIG. 5, when the ship is navigating in a narrow river channel or canal, there is a risk of collision everywhere around the hull. Therefore, it can be assigned to the recognition areas Z1 to Z6 in the order of the left bow, the right bow, the right side of the hull, the right stern, the left stern, and the left side of the hull. Thereby, the direction from which the obstacle approaches can be accurately detected. It should be noted that the area division step described in the present invention may be preset, or the user may select different patterns of area division.

[0025] As shown in FIGS. 1 and 2, the obstacle recognition step S2 takes the moving carrier V as the center, captures an Around View Monitor (AVM) image, and recognizes the obstacle B in this image. In this embodiment, image segmentation is performed on the around view image to position the position and its edge of the object in the image. For example, after enlarging and shrinking the around view image to a custom size, all pixels in the around view image are classified by inputting them into a semantic segmentation model, and it is determined whether each pixel is the obstacle B. After classifying the pixels P belonging to the obstacle B, a plurality of pixels P located on the edge of the obstacle B can be found by a contour tracing algorithm.

[0026] Also, the obstacle B is distributed in a plurality of recognition areas Z. Depending on the way it is assigned to the plurality of recognition areas Z, the obstacle B may be detected simultaneously in a plurality of adjacent recognition areas Z. Therefore, a plurality of pixels P belonging to the same obstacle B can be distributed in different recognition areas Z, and different calculation results may be obtained when detecting the same obstacle B in the plurality of recognition areas Z respectively.

[0027] The distance calculation step S3 calculates the distance between each obstacle B in each recognition area Z and the border R corresponding to the recognition area Z, that is, the relative distance D between the obstacle B detected in each recognition area Z and the moving carrier V. In this embodiment, since the around view image is enlarged and shrunk to a custom size, the actual distance between pixels can be converted. Also, the distance calculation step S3 can install a plurality of sampling points M at equal intervals on each border R. Thereby, the distances between the plurality of pixels P of the obstacle B and the plurality of sampling points M are calculated, and the shortest one among the obtained results is taken as the relative distance D. During the process of the mobile carrier V moving, the distance calculation step S3 continuously calculates a plurality of relative distances D for detecting the obstacle B in a plurality of recognition areas Z. It should be noted that in the present invention, the shortest distance is used as the relative distance D. However, those skilled in the art can understand that other distances can also be defined as the relative distance D. Therefore, the relative distance D in the present invention is not limited to the shortest distance.

[0028] The plurality of sampling points M may be installed at equal intervals on a plurality of borders R with different lengths. As a result, the border R with a shorter length has fewer sampling points M. Also, each border R may have the same number of sampling points M. As a result, the sampling points M on the border R with a shorter length have a higher distribution density. However, the number and distribution of the sampling points M in the present invention are not limited.

[0029] Also, the distance calculation step S3 can set a sampling period, an interval time, and a warning threshold. As a result, a plurality of relative distances D are calculated at regular intervals during the sampling period, and the plurality of relative distances D have the warning threshold as the warning lower limit. Note that the sampling period may be 0.3 to 0.5 seconds, the interval time may be 0.03 to 0.05 seconds, and the warning threshold is preferably 1 meter.

[0030] The distance calculation step S3 can provide a collision warning based on the relationship between the calculation results of the plurality of relative distances D and the warning threshold. Taking an example for explanation, in the plurality of relative distances D calculated during the sampling period, when the data that is less than or equal to the warning threshold accounts for more than the risk ratio, a collision warning can be issued. Note that the risk ratio is preferably 60% to 80%. Specifically, if the sampling period is 0.3 seconds and the interval time is 0.03 seconds, 10 relative distances D are taken. Among them, if 7 relative distances D are less than or equal to 1 meter, and the risk ratio is set to 60% or 70%, a warning is issued. If the risk ratio is set to 80%, it is determined to be safe.

[0031] In other embodiments, an average value can be calculated for a plurality of relative distances D calculated within the sampling period, and a collision warning can be issued when the average value is less than or equal to the warning threshold value.

[0032] Also, in other embodiments, a collision warning can be issued when the change rate of the relative distance D with respect to time is less than or equal to -1 m / s. The above change rate is the ratio of the difference between the numerical values of consecutive relative distances D and the interval time. Specifically, when the results of calculating consecutive relative distances D are 5 meters and 4.95 meters respectively, and the interval time is 0.04 seconds, the change rate is -1.25 m / s, which means that the speed at which the moving carrier V approaches an object that can become an obstacle is higher than the safe range. Therefore, a warning is issued, and a reaction time for avoiding the obstacle can be ensured in advance.

[0033] In summary, the method for detecting an underwater obstacle of the present invention divides the surrounding detection range into areas, and further classifies the obstacles by using the around-view image, so as to detect a plurality of obstacles respectively by using a plurality of recognition areas, or issue a warning for the same obstacle, thereby reducing the complexity of the calculation in distance measurement. Furthermore, based on different parts and angles of the main body of the moving carrier, the distance to the obstacle is calculated, so that the accuracy of detection and distance measurement can be improved, and the effect of providing collision warnings at multiple points can be achieved.

[0034] The present invention is disclosed with the above embodiments, but they do not limit the present invention. Those skilled in the art may make various changes to the above embodiments without departing from the spirit and scope of the present invention, and still fall within the scope protected by the present invention. Therefore, the scope protected by the present invention includes all changes within the scope of the language described in the claims hereinafter and all equivalent scopes.

Explanation of Reference Signs

[0035] S1 ··· Area Division Step S2 ··· Obstacle Recognition Step S3 ··· Distance Calculation Step V ··· Moving Carrier B ··· Obstacle R ··· Border Z, Z1, Z2, Z3, Z4, Z5, Z6 ··· Recognition Areas P ··· Pixel M ··· Sampling Point D ··· Relative Distance

Claims

1. a dividing step of dividing the outline of the body of the moving carrier in the two-dimensional horizontal plane into a plurality of continuous borders, and extending from each of the borders to the outside of the moving carrier as a recognition area, the plurality of recognition areas being continuous and surrounding the periphery of the moving carrier; an obstacle recognition step of capturing an around-view image with the mobile carrier as a center and recognizing at least one obstacle distributed in the plurality of recognition areas in the around-view image; A method for detecting obstacles on water, comprising: a distance calculation step of calculating the distance between each obstacle in each of the recognition areas and the border to which the recognition area corresponds, i.e., the relative distance between an obstacle detected in each of the recognition areas and the mobile carrier, and continuously calculating multiple relative distances detected in multiple recognition areas during the movement of the mobile carrier.

2. The method of claim 1, further comprising utilizing a semantic segmentation model to classify all pixels in the around-view image and classify pixels belonging to a plurality of the obstacles.

3. 3. The method for detecting obstacles on water as described in claim 2, characterized in that each of the borders has a plurality of equally spaced sampling points, and each of the relative distances is the shortest of all distances sampled between a plurality of pixels of each of the obstacles and each of the plurality of sampling points.

4. A method for detecting water obstacles as described in claim 1, characterized in that in the distance calculation step, a sampling period, an interval time and an alert threshold are set, a plurality of the relative distances are calculated with a constant interval time within the sampling period, and the alert threshold is set as the lower alert limit for the plurality of the relative distances.

5. the sampling period is between 0.3 and 0.5 seconds; The interval time is 0.03 to 0.05 seconds, 5. The method for detecting obstacles on water according to claim 4, wherein the alert threshold is 1 meter.

6. The method for detecting obstacles on water as described in claim 4, characterized in that a collision warning is issued when data that is smaller than or equal to the alert threshold for multiple relative distances within the sampling period is equal to or greater than a danger rate.

7. 7. The method for detecting an obstacle on water according to claim 6, wherein the danger rate is 60% to 80%.

8. 5. The method for detecting an obstacle on water according to claim 4, further comprising issuing a collision warning when an average value of a plurality of said relative distances within said sampling period is smaller than said alert threshold or equal to said alert threshold.

9. 5. The method for detecting obstacles on water as claimed in claim 4, further comprising issuing a collision warning when a rate of change of the plurality of relative distances with respect to time is less than or equal to -1 meter / second.

Images