A method for calculating three-dimensional target bounding boxes in automatic parking

Abstract

The present invention discloses a method for calculating a three-dimensional target box in automatic parking, which includes the following steps: Step 1: Obtain the target data sent by the vehicle-mounted sensor and parse the target data; Step 2: Based on the data obtained by parsing, determine whether the currently parsed frame of target data is the first frame of data of the target; if so, proceed to Step 4; if not, proceed to Step 3; Step 3: Weight the currently obtained frame of target data in Step 2 with the historical data saved in the previous frame to obtain historical weighted data; Step 4: Calculate the three-dimensional box of the target based on the first frame of data obtained in Step 2 or the historical weighted data obtained in Step 3; Step 5: Output the three-dimensional box of the target to the vehicle-mounted controller, and loop through Steps 1 to 5. The method of the present invention can output a stable three-dimensional target box to the vehicle-mounted controller in real time, and the output three-dimensional target box has less jitter, which is more conducive to the vehicle-mounted controller for path planning.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic parking, and particularly relates to a method for calculating a three-dimensional target box in automatic parking. Background Art

[0002] Automatic parking is to achieve automatic recognition of available parking spaces and automatically and accurately park the vehicle into the parking space without manual intervention, through in-vehicle sensors and in-vehicle processors. To achieve automatic parking, the in-vehicle processor needs to precisely control the vehicle. To achieve precise control, it is first necessary to accurately perceive the surrounding environment. In addition to perceiving the parking space, it is also necessary to accurately perceive the information of surrounding targets, including information such as stationary vehicles, moving vehicles, and pedestrians. Therefore, to achieve better automatic parking, it is necessary to more accurately identify the target information, including the identification of the target box. The prior art has defects in the identification of the target box. For example, Patent CN201811591235.7 discloses a method for converting a two-dimensional bounding box of an object into a three-dimensional position of an autonomous vehicle, which only calculates the target data based on the current frame. However, since the in-vehicle sensor cannot achieve absolute precision in identifying the target, there will be a certain error if only calculating the target information based on the current frame data of the sensor. Summary of the Invention

[0003] Aiming at the above deficiencies existing in the prior art, the purpose of the present invention is to provide a method for calculating a three-dimensional target box in automatic parking, so as to solve the problems of unstable identification of the target box, large error fluctuations, and low accuracy in the prior art.

[0004] To solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A method for calculating a three-dimensional target box in automatic parking, comprising the following steps:

[0006] Step 1: Obtain the target data sent by the in-vehicle sensor and parse the target data;

[0007] Step 2: Based on the data parsed in Step 1, determine whether the current frame target data parsed is the first frame data of the target; if so, proceed to Step 4; if not, proceed to Step 3;

[0008] Step 3: Weight the current frame target data obtained in Step 2 with the historical data saved in the previous frame to obtain historical weighted data;

[0009] Step 4: Calculate the three-dimensional box of the target based on the first frame data obtained in Step 2, or the historical weighted data obtained in Step 3;

[0010] Step 5: Output the three-dimensional bounding box of the target to the vehicle-mounted controller, and loop through Steps 1-5.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] The method of the present invention can utilize the data sent in real time by the sensors and, in combination with historical data, calculate in real time the three-dimensional coordinate bounding boxes of targets including vehicles, pedestrians, etc., and transmit them to the vehicle-mounted controller for vehicle control; the method of the present invention can output stable three-dimensional target bounding boxes to the vehicle-mounted controller in real time. Compared with the prior art that only determines the target bounding box based on the current frame, the present invention has better stability, with less jitter in the output three-dimensional target bounding boxes, which is more conducive to the vehicle-mounted controller for path planning, thus making automatic parking safer. Description of the Drawings

[0013] Figure 1 It is a flowchart of a method for calculating a three-dimensional target bounding box in an automatic parking of the present invention. Detailed Embodiments

[0014] In order to make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the present invention will be further described in detail below with reference to the drawings. The described embodiments should not be construed as limiting the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0015] In automatic parking, for the vehicle to safely park in the parking space, it is necessary to effectively avoid stationary or moving vehicles and passing pedestrians. The controller needs to obtain stable three-dimensional target bounding box information to accurately control the vehicle so that the vehicle can avoid the target. The present invention provides a method for calculating a three-dimensional target bounding box in automatic parking, including the following steps:

[0016] Step 1: Obtain the target data sent by the vehicle-mounted sensors and parse the target data. Among them, the target data is sent in real time by the vehicle-mounted sensors and is parsed in real time during vehicle-mounted processing. The data includes the length, width, height, angle, ordinate point, and abscissa point of the target. The three-dimensional coordinate system of the data has the center of the front bumper of the vehicle as the origin, the due front of the vehicle as the x-axis, the due left as the y-axis, and the due top as the z-axis.

[0017] Step 2: Based on the data parsed in Step 1, determine whether the currently parsed target data of the current frame is the first frame data of the target; if so, it indicates that there is no historical data frame for the target, then go to Step 4, and after Step 4 is completed, save the currently parsed target data of the current frame as historical data. If not, go to Step 3.

[0018] Step 3: Weight the current frame target data obtained in Step 2 with the historical data saved in the previous frame to obtain the historical weighted data. Specifically, in the current frame target data, the length, height, and width of the target are weighted and summed with the historical data saved in the previous frame to obtain the historical weighted data.

[0019] The calculation method of the historical weighted data is as follows:

[0020] Historical weighted data = Historical data × λ + Current data × (1 - λ), where 0 < λ < 1. The specific value of λ can be obtained through multiple experiments to get the optimal value.

[0021] Step 4: Calculate the 3D bounding box of the target based on the first frame data obtained in Step 2 or the historical weighted data obtained in Step 3. The 3D bounding box of the target is calculated through the following steps:

[0022] (1) Determine the 3D model of the target according to the length, width, and height of the target in the calculated historical weighted data or the first frame data.

[0023] (2) Calculate the 2D top-down coordinates of 4 points of the target according to the length, width, angle, abscissa, and ordinate of the target in the data.

[0024] (3) Calculate the 3D coordinates of 8 points of the target based on the 2D top-down coordinates obtained in step (2) and the height information of the target in the data, so as to obtain the 3D bounding box of the target.

[0025] Step 5: Output the 3D bounding box of the target to the vehicle-mounted controller, and loop through Steps 1 - 5. Since the position of the vehicle is changing in real time during the automatic parking process, the output 3D bounding box of the target also changes in real time. This enables the vehicle to monitor surrounding vehicles, pedestrians, and other targets in real time during the automatic parking process, ensuring that the 3D bounding box of the target can be output in real time, and further enabling more precise control of the vehicle.

[0026] Example:

[0027] When testing the method of the present invention, two vehicles are required. One vehicle is used as the host vehicle, and the other vehicle is used as the target vehicle. One or more sensors to be tested, such as a front-view camera, a rear-view camera, a millimeter-wave radar, an ultrasonic radar, a lidar, etc., are installed on the host vehicle. The test scenario is an underground garage with obvious parking lot lines, and there are other vehicles parked in some parking spaces in the garage. After preparation, the two vehicles conduct following vehicle tests, oncoming driving tests, the target vehicle stops on the lane, the host vehicle bypasses the target vehicle test, and a passing test with pedestrians in front in the underground garage. During the test, the 3D coordinate bounding boxes of vehicle and pedestrian targets are calculated in real time by combining the data transmitted by the sensors in real time with the historical data, and are transmitted to the vehicle-mounted controller for vehicle control.

[0028] In the experiment, the three-dimensional object bounding box is detected according to the proposed detection method. As Figure 1 shown, the method of the present invention is divided into five steps in total. The first step is to receive and parse the target data. First, the target data transmitted by the vehicle-mounted sensor is received in real time and parsed in the vehicle-mounted processor. The data includes the length, width, height, angle, vertical coordinate point, horizontal coordinate point, etc. of the target. The three-dimensional coordinate system of the data is that the center of the front bumper of the vehicle is used as the origin, the due front of the vehicle is used as the x-axis, the due left of the vehicle is used as the y-axis, and the due top of the vehicle is used as the z-axis. If the target is a vehicle, the output horizontal and vertical coordinates are based on the center of the rear bumper of the target vehicle as the target point. If the target is a pedestrian, the horizontal and vertical coordinates are based on the center of the pedestrian as the target point.

[0029] The second step is to determine whether it is the first frame. Determine whether the target data of the currently received and parsed frame is the first frame data of the target. If it is, then there is no historical data frame for this target, and directly jump to the fourth step. Use the currently parsed data to calculate the three-dimensional object bounding box of this target, and save the data of this frame as historical data. If it is not the first frame data, then proceed to the next step. If there is no continuous interruption in the data of the same vehicle, then the second segment of data is considered a new target.

[0030] The third step is weighted summation. The length, height, and width of the target in the target data of the current frame are weighted and summed with the historical data saved in the previous frame to obtain the historical weighted data, that is, historical weighted data = historical data × λ + current data × (1 - λ), where 0 < λ < 1, and the specific value of λ can be experimented multiple times to obtain an optimal solution.

[0031] The fourth step is to calculate the three-dimensional object bounding box. First, determine the three-dimensional model of the target according to the length, width, and height of the target in the calculated historical weighted data or the first frame data. Then, calculate the two-dimensional top-down coordinates of the 4 points of this target according to the length, width, angle, horizontal coordinate, and vertical coordinate in the data. Finally, calculate the three-dimensional coordinates of the 8 points of the target according to the two-dimensional coordinates and the height information of the target in the data, so as to obtain the three-dimensional bounding box of the target.

[0032] The fifth step is to output and loop. Output the obtained three-dimensional object bounding box to the vehicle-mounted controller.

[0033] Loop through the above five steps in sequence to output a stable three-dimensional object bounding box to the vehicle-mounted controller in real time. Compared with the traditional method that only relies on the current frame, the method of the present invention has better stability, the jitter of the output three-dimensional object bounding box is smaller, and it is more conducive to the vehicle-mounted controller to perform path planning for automatic parking.

[0034] As described above, the reminder system of the present invention is not limited to the above configuration, and other systems that can implement the embodiments of the present invention can fall within the scope protected by the present invention.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Those of ordinary skill in the art should understand that any modifications or equivalent replacements made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for calculating a three-dimensional target box in automatic parking, characterized in that, It includes the following steps: Step 1: Obtain the target data sent by the vehicle-mounted sensor and parse the target data; Step 2: Based on the data parsed in Step 1, determine whether the currently parsed frame of target data is the first frame of data for the target; if so, proceed to Step 4; if not, proceed to Step 3; Step 3: Weight the currently parsed frame of target data obtained in Step 2 with the historical data saved in the previous frame to obtain the historical weighted data; in Step 3, in the currently parsed frame of target data, weight and sum the length, height, and width of the target with the historical data saved in the previous frame to obtain the historical weighted data; the calculation method of the historical weighted data is as follows: Historical weighted data = historical data × λ + current data × (1 - λ), where 0 < λ < 1; Step 4: Calculate the three-dimensional bounding box of the target based on the first frame of data obtained in Step 2 or the historical weighted data obtained in Step 3; in Step 4, calculate the three-dimensional bounding box of the target through the following steps: (1) Determine the three-dimensional model of the target based on the length, width, and height of the target in the calculated historical weighted data or the first frame of data; (2) Calculate the two-dimensional top-down coordinates of the 4 points of the target based on the length, width, angle, abscissa, and ordinate of the target in the data; (3) Calculate the three-dimensional coordinates of the 8 points of the target based on the two-dimensional top-down coordinates obtained in Step (2) and the height information of the target in the data, so as to obtain the three-dimensional bounding box of the target; Step 5: Output the three-dimensional bounding box of the target to the vehicle-mounted controller and loop through Steps 1 to 5.

2. The calculation method of the three-dimensional target box in automatic parking according to claim 1, wherein In Step 1, the three-dimensional coordinate system of the data has the center of the front bumper of the vehicle as the origin, the due front of the vehicle as the x-axis, the due left as the y-axis, and the due top as the z-axis.

3. The calculation method of the three-dimensional target box in automatic parking according to claim 1, wherein In Step 2, if the currently parsed frame of target data is the first frame of data, after Step 4 is completed, save the currently parsed frame of target data as the historical data.

4. The calculation method of the three-dimensional target box in automatic parking according to claim 1, wherein In Step 1, the data includes the length, width, height, angle, ordinate point, and abscissa point of the target.

Images