A parking space tracking method for intelligent driving

Abstract

The present invention relates to a parking space tracking method for intelligent driving, comprising: initializing tracking location information; correcting the initialized tracking location based on location information collected in real time by sensors; and performing a secondary correction on the corrected tracking location based on prior shape constraints to obtain accurate and stable location information. After initializing the location information, the method utilizes measured location information input by multiple sensors to iteratively update the tracking location position information, outputting accurate and stable location information, thereby assisting downstream modules in fully automated precision parking.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent driving technology, and in particular relates to a parking space tracking method for intelligent driving. Background Art

[0002] Currently, when smart cars perform automatic parking in scenarios like ports and public parking lots, they are unable to directly park when a parking space is identified due to the vehicle's posture. To achieve efficient, fully automated parking, the vehicle's posture must be adjusted based on the spatial context of the scene. During this adjustment period, due to limitations in parking space detection accuracy or vehicle control precision, it's impossible to guarantee stable and accurate parking space information all at once. Furthermore, existing parking space tracking algorithms often rely on visual location detection. If visual detection is unstable, the fused location will also experience certain position fluctuations, potentially causing the vehicle to stray from its parking space during automated parking, posing a certain risk. Summary of the Invention

[0003] The purpose of the present invention is to address the problems existing in the prior art and propose a parking space fusion tracking method based on multi-sensor input, which can achieve highly efficient fully automatic parking.

[0004] In order to achieve the above objectives, the present invention provides a parking space tracking method for intelligent driving, comprising the following steps:

[0005] Step 1: Initialize tracking location information;

[0006] Step 2: Modify the initialized tracking location based on the location information collected by the sensor in real time;

[0007] Step 3: Perform secondary correction on the corrected tracking location according to the prior shape constraint to obtain accurate and stable location information.

[0008] The present invention further adopts the following technical solution:

[0009] In step 1, the tracking location is initialized based on the map location information, and the prior position information and prior shape information of the tracking location are given. The tracking location is rectangular in shape, and the prior position information of the tracking location is described by the coordinates of the four corner points of the rectangle (p0, p1, p2, p3), where p0 and p3 represent the entry points when parking, and p1 and p2 represent the bottom points of the location. Each corner point has a longitudinal distance x and a lateral distance y.

[0010] In step 2, multiple sensors that can be used to input parking location information during parking include a rear-view camera installed at the center of the rear of the vehicle and two side cameras installed on the left and right sides of the front of the vehicle and facing backward.

[0011] In step 2, multi-frame fusion tracking is first performed based on the storage location information collected by the sensor to obtain processed image information. Then, the tracked storage location is corrected based on the processed image information, and the storage location position information measured by the sensor is described by the coordinates of four corner points (m0, m1, m2, m3). Among them, m0 and m3 represent the entry points when parking, and m1 and m2 represent the bottom points of the storage location. The initial tracking storage location corner points are then corrected by weighted averaging. The calculation formula is as follows:

[0012] p′ k =w k *p′ k +(1-w k )*m k

[0013] Where p′ k represents the kth tracking location corner point after correction, w k Represents the confidence of the kth tracking location corner point detected by the camera, m k Represents the coordinate information of the kth tracking location corner point measured by the sensor, k represents the index of the location corner point and k = 0, 1, 2, 3, w represents the confidence level of the corner point detected by the camera, and the value range is a number between [0, 1]. It can be output by vision or set to an empirical value during testing, such as 0.7.

[0014] In step 3, when verifying the corrected tracking parking space, based on the prior shape constraints, the diagonal lengths of the rectangle are equal, and the length and width of the actual parking space remain unchanged, the shape errors of the four corner points are calculated and the corner point with the largest shape error is selected. Then, the remaining three corner points except the corner point with the largest shape error are used to generate a new parking location.

[0015] Preferably, in step 3, the shape error of the corner point p0 is calculated as follows: error(p0)=distance(p0,p1)+distance(p0,p2)+distance(p0,p3)-LWd;

[0016] The shape error of the corner point p1 is calculated as follows:

[0017] error(p1)=distance(p1,p0)+distance(p3,p1)+distance(p3,p2)-LWd;

[0018] The shape error of the corner point p2 is calculated as follows:

[0019] error(p2)=distance(p2,p0)+distance(p2,p1)+distance(p2,p3)-LWd;

[0020] The shape error of the corner point p3 is calculated as follows:

[0021] error(p3)=distance(p3,p0)+distance(p0,p2)+distance(p0,p3)-LWd;

[0022] Where L is the length of the initial storage location, W is the width of the initial storage location, and d is the diagonal length of the initial storage location.

[0023] In step 3, three corner points are used to generate new library sites, specifically in the following manner:

[0024] Assume that the equations of the lines of the four sides of the rectangle are y=kx+b. Two of the three corner points belong to the long side of the storage location, and two belong to the short side of the storage location. Then, the slope of the long side and the slope of the short side of the storage location can be calculated based on the information of the storage location.

[0025] Since the two opposite sides of the rectangle are parallel, the slopes of the long and short sides and the positions of the three corner points are known, and the straight line equation of each side can be calculated. The new storage location corner point is the intersection of the straight line equations of the corresponding two long and short sides.

[0026] The parking space fusion tracking method of the present invention uses map location information to determine the target location's prior location and shape information. It then performs multi-frame fusion tracking based on the location information input by sensors. The initial location is corrected, and then the location's shape information is used to perform a secondary correction to align it with the actual location information. Finally, the method outputs stable and accurate location information to downstream users. In summary, the present invention combines detection information from multiple sensors to continuously correct parking spaces during the parking process, outputting stable and accurate location information to achieve highly efficient, fully automated parking.

[0027] After initializing the storage location information, the method of the present invention uses the measured storage location information input by multiple sensors to iteratively update the tracking storage location position information to output accurate and stable storage location information, thereby assisting downstream modules in fully automatic and precise parking.

[0028] The method of the present invention can be applied to the research and development of fully automatic parking systems for intelligent vehicles, thereby promoting the intelligent development of parking. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be further described below with reference to the accompanying drawings.

[0030] Figure 1This is a flow chart of the parking space fusion tracking method of the present invention.

[0031] Figure 2 Schematic diagram of the storage location corner points in the present invention.

[0032] Figure 3 This is a schematic diagram of the correction of storage location corner points in the present invention. DETAILED DESCRIPTION

[0033] Example 1

[0034] like Figure 1 As shown, a parking space tracking method for intelligent driving includes the following steps:

[0035] Step 1: Initialize tracking location information.

[0036] Initialize the tracking location according to the map location information, and give the prior position information and prior shape information of the tracking location. This embodiment takes the rectangular tracking location as an example, and uses the coordinates of the four corner points (p0, p1, p2, p3) of the rectangular location to describe the prior position information of the tracking location (see Figure 2 ), where p0 and p3 represent the parking entry points, p1 and p2 represent the bottom points of the parking space, and each corner point contains the longitudinal distance x and the lateral distance y.

[0037] Step 2: Correct the initialized tracking location based on the location information collected in real time by multiple sensors. The sensors used are a rearview camera mounted at the center of the rear of the vehicle and two side cameras mounted on the left and right sides of the front of the vehicle and facing backward.

[0038] The method for correcting the initial storage location is as follows: First, multi-frame fusion tracking is performed based on the storage location information collected by the sensor to obtain processed image information. Then, the tracked storage location is corrected based on the processed image information. The coordinates of the four corner points (m0, m1, m2, m3) are used to describe the visually detected storage location information. m0 and m3 represent the parking entry points, and m1 and m2 represent the bottom points of the storage location. The initial tracking location corner points are then corrected using a weighted average method. The calculation formula is as follows:

[0039] p′ k =w k *p′ k +(1-w k )*m k

[0040] Where p′ k represents the kth tracking location corner point after correction, w k Represents the confidence of the kth tracking location corner point detected by the camera, m kRepresents the coordinate information of the kth tracking location corner point measured by the sensor, k represents the index of the location corner point and k = 0, 1, 2, 3, w represents the confidence level of the corner point detected by the camera, and the value range is a number between [0, 1]. It can be output by vision or set to an empirical value during testing, such as 0.7.

[0041] Step 3: Perform secondary correction on the corrected tracking location according to the prior shape constraint to obtain accurate and stable location information and output it to the downstream module.

[0042] When verifying the corrected tracking parking location, based on the prior shape constraints, the diagonal lengths of the rectangle are equal, and the length and width of the actual parking location remain unchanged, the shape errors of the four corner points are calculated and the corner point with the largest shape error is selected. Then, the remaining three corner points except the corner point with the largest shape error are used to generate a new parking location.

[0043] Specifically, if Figure 3 As shown, the shape error of the corner point p2 is calculated as follows:

[0044] error(p2)=distance(p2,p0)+distance(p2,p1)+distance(p2,p3)-LWd

[0045] The shape error of the corner point p0 is calculated as follows:

[0046] error(p0)=distance(p0,p1)+distance(p0,p2)+distance(p0,p3)-LWd

[0047] The shape error of the corner point p1 is calculated as follows:

[0048] error(p1)=distance(p1,p0)+distance(p3,p1)+distance(p3,p2)-LWd

[0049] The shape error of the corner point p3 is calculated as follows:

[0050] error(p3)=distance(p3,p0)+distance(p0,p2)+distance(p0,p3)-LWd

[0051] Where L is the length of the initial storage location, W is the width of the initial storage location, and d is the diagonal length of the initial storage location.

[0052] The method of generating new library sites using three corner points is: let the equations of the straight lines of the four sides of the rectangle be y=kx+b, and b0=y0-k01 If two of the x0 lines belong to the long side of the location and two belong to the short side of the location, the slopes of the long and short sides of the location can be calculated based on the location information. If the two opposite sides of a rectangle are parallel, and the slopes of the long and short sides are known, as well as the locations of the three corner points, the equations of the lines for each side can be calculated. The new corner point of the location is the intersection of the equations of the lines for the two long and short sides.

[0053] Take the generation of a new corner point p2 as an example:

[0054] Given p0(x0,y0), p1(x1,y1), p3(x3,y3), the equation of the four sides is y=k i x+b i (i represents the line index)

[0055] Long side slope k 32 =k 01 =(y1-y0) / (x1-x0)

[0056] Short side slope k 12 =k 03 =(y3-y0) / (x3-x0)

[0057] The straight line y=k passing through the corner point 0,1 01 x+b0,b0=y0-k 01 *x0

[0058] The straight line y=k passing through the corner point 0,3 03 x+b1,b1=y0-k 03 *x0

[0059] The straight line y=k passing through the corner points 1 and 2 12 x+b2,b2=y1-k 12 *x1

[0060] The straight line y=k passing through the corner points 2 and 3 32 x+b3,b3=y3-k 32 *x3

[0061] Finally calculate the straight line y=k 12 x + b2 and y = k 32 The intersection of x+b3 gives the new corner point p2.

[0062] It should be noted that the execution order of the above steps is determined by their internal logic and functions. As long as the order of execution can achieve the desired results of the technical solution disclosed in this patent, it should not impose any restrictions or constraints on the implementation of the present invention and its embodiments. In addition to the above embodiments, the present invention can also have other implementation methods. Any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of protection required by the present invention.

Claims

1. A parking space tracking method for intelligent driving, characterized in that: The following steps are involved: Step 1: Initialize the tracking location information. Initialize the tracking location according to the map location information and provide the prior location information and shape information of the tracking location. The shape of the tracking location is rectangular, and the coordinates of the four corner points of the rectangle (p0, p1, p2, p3) are used to describe the prior location information of the tracking location. P0 and p3 represent the entry points when parking, and p1 and p2 represent the bottom points of the location. Each corner point contains the longitudinal distance x and the lateral distance y. Step 2: Correct the initialized tracking location based on the location information collected by the sensor in real time. First, perform multi-frame fusion tracking based on the location information collected by the sensor to obtain processed image information. Then, correct the tracking location based on the processed image information. The location information measured by the sensor is described by the coordinates of the four corner points (m0, m1, m2, m3). Then, the initial tracking location corner points are corrected by weighted averaging. The calculation formula is as follows: p′ k =w k *p′ k +(1-w k )*m k Where p′ k represents the kth tracking location corner point after correction, w k Represents the confidence of the kth tracking location corner point detected by the camera, m k Represents the coordinate information of the kth tracking location corner point measured by the sensor, k represents the index of the location corner point and k = 0, 1, 2, 3; Step 3: Perform secondary correction on the corrected tracking location according to the prior shape constraints to obtain accurate and stable location information. When verifying the corrected tracking location, calculate the shape errors of the four corner points according to the prior shape constraints, the diagonal lengths of the rectangle are equal, and the length and width of the actual parking location remain unchanged. Select the corner point with the largest shape error, and then use the remaining three corner points except the corner point with the largest shape error to generate a new location.

2. The parking space tracking method for intelligent driving according to claim 1, characterized in that: In step 2, multiple sensors may be used to input parking location information when parking, including a rear-view camera installed at the center of the rear of the vehicle and two side cameras installed on the left and right sides of the front of the vehicle and facing rearward.

3. The parking space tracking method for intelligent driving according to claim 1, characterized in that: In step 3, the shape error of the corner point p0 is calculated as follows: error(p0)=distance(p0,p1)+distance(p0,p2)+distance(p0,p3)-LWd; the shape error of the corner point p1 is calculated as follows: error(p1)=distance(p1,p0)+distance(p3,p1)+distance(p3,p2)-LWd; The shape error of the corner point p2 is calculated as follows: error(p2)=distance(p2,p0)+distance(p2,p1)+distance(p2,p3)-LWd; the shape error of corner point p3 is calculated as follows: error(p3)=distance(p3,p0)+distance(p0,p2)+distance(p0,p3)-LWd; Where L is the length of the initial storage location, W is the width of the initial storage location, and d is the diagonal length of the initial storage location.

4. The parking space tracking method for intelligent driving according to claim 1, characterized in that: In step 3, three corner points are used to generate new library sites, specifically in the following manner: Assume that the equations of the lines of the four sides of the rectangle are y=kx+b. Two of the three corner points belong to the long side of the storage location, and two belong to the short side of the storage location. Then, the slopes of the long side and the short side of the storage location can be calculated based on the information of the storage location. Since the two opposite sides of the rectangle are parallel, the slopes of the long and short sides and the positions of the three corner points are known, and the straight line equation of each side can be calculated. The new storage location corner point is the intersection of the straight line equations of the corresponding two long and short sides.

Images