METHOD OF DETERMINING A GROUND PLANE ON THE BASIS OF A DEPTH IMAGE

• US 20150036887A1
• Filed: 03/15/2013
• Published: 02/05/2015
• Est. Priority Date: 03/23/2012
• Status: Active Grant

• Current Assignee: Commissariat a L'Energie Atomique (French Alternative Energies and Atomic Energy Commission)
• Original Assignee: Commissariat a L'Energie Atomique (French Alternative Energies and Atomic Energy Commission)
• Inventors: Allezard, Nicolas, Lucat, Laurent, Hervieu, Alexandre
• Granted Patent: US 9,361,696 B2
• Time in Patent Office: Days
• US Class Current: 382/106
• CPC Class Codes: G06T 2207/10021 Stereoscopic video; Stereoscopic image sequence
• CPC Class Codes: G06T 2207/10028 Range image; Depth image; 3D point clouds
• CPC Class Codes: G06T 2207/30252 Vehicle exterior; Vicinity of vehicle
• CPC Class Codes: G06T 7/50 Depth or shape recovery
• CPC Class Codes: G06T 7/75 involving models
• CPC Class Codes: G06V 20/64 Three-dimensional objects

Abstract

A method of determining a triplet of parameters defining a ground plane based on a depth image includes determining a plurality of triplets of parameters, each defining a ground plane, spatial filtering of the parameters, and temporal filtering of the parameters. The temporal filtering is dependent on an indicator of quality of depth image data available for determination of the ground plane. The temporal filtering is dependent on an indicator of quality of the ground plane determined after spatial filtering. Similarly, the spatial filtering can be parameterized as a function of the two indicators. Globally, the more the estimated plane for a depth image can “explain” the points in this image, the more accurately planes of the subsequent images are determined. For the spatial filtering, the ground plane is searched in a more limited space, and, for the temporal filtering, the previously estimated ground planes considered to a lesser degree.

Claims

1. A method of determining a triplet of parameters defining a ground plane in a three-dimensional reference frame on the basis of a depth image obtained from a sequence of images, comprising:

selecting valid points from the current depth image, the valid points being the points of the image whose distance from a point of origin of the three-dimensional frame of reference is in the range between a predetermined minimum distance and a predetermined maximum distance,

determining an indicator R_3d for the current depth image, said indicator being dependent on the number of valid points in the current depth image, and on the total number of points in this image,

selecting a set of points of interest in the current depth image, each point of interest being selected if a distance between said point of interest and the ground plane determined for a previous image in the sequence is below a threshold d_sel,

determining a triplet of spatially filtered parameters on the basis of points of interest, and

determining a triplet of temporally filtered parameters as a function of the indicator R_3d determined for the current depth image or for a previous image in the sequence.

2. The method as claimed in claim 1, further comprising determining an indicator R_pl for the current depth image, said indicator being dependent on the number of points in said image whose distance from the ground plane defined by the triplet of spatially filtered parameters is less than a predetermined distance d_p, and on the number of valid points in this image, the determining the triplet of temporally filtered parameters also being dependent on the indicator R_pl determined for the current depth image or for a previous image in the sequence.

3. The method as claimed in claim 2, wherein the determining the triplet of temporally filtered parameters includes determining an innovation parameter K, said parameter being:

equal to zero if the indicator R_3d determined for the current depth image or for a previous image in the sequence is below a predetermined threshold S_3d, or if the indicator R_pl determined for the current depth image or for a previous image in the sequence is below a predetermined threshold S_pl,

equal to the value of the indicator R_pl determined for the current depth image or for a previous image in the sequence, less the value of the predetermined threshold S_pl otherwise,

each temporally filtered parameter V_t being determined as a function of the innovation parameter K, of the temporally filtered parameter determined for the previous image in the sequence V_t-1, and of the spatially filtered parameter determined for the current depth image in the sequence V_s.

4. The method as claimed in claim 3, wherein each temporally filtered parameter V_t is determined by the relation:

V_t=V_t-1·

(1−

K)+V_s·

K.

5. The method as claimed in claim 2, wherein the determining a triplet of spatially filtered parameters comprises:

determining a plurality of triplets of parameters on the basis of the set of points of interest, and

selecting, for each type of parameter, parameters whose value lies within a parameterable validity range, the validity range of each type of parameter being centered on the value of this parameter determined for a previous image in the sequence, and its width being determined as a function of the indicator R_pl determined for a previous image in the sequence, and of the indicator R_3d determined for the current image or for a previous image in the sequence.

6. The method as claimed in claim 5, wherein the determining a plurality of triplets of parameters comprises:

pseudorandom determination of a plurality of triplets of points in the set of points of interest, and

determination, for each triplet of points, of a triplet of parameters defining a ground plane passing through the three points.

7. The method as claimed in claim 5, wherein the determining a triplet of spatially filtered parameters further comprises:

determining, for each type of parameter, an average parameter whose value is equal to an average value of the selected parameters, said average parameters defining an average ground plane P_m,

of selecting a set of calculation points from among the points of interest, the calculation points being the points of interest whose distance from the average ground plane P_m is less than a predetermined distance d_g, and

determining the triplet of spatially filtered parameters on the basis of the calculation points.

8. The method as claimed in claim 7, wherein the triplet of spatially filtered parameters is determined by the least squares method applied to the distances between the calculation points and the mean ground plane P_m.

9. The method as claimed in claim 5, wherein the width of the validity range of each parameter is:

increased if the indicator R_pl determined for a previous image in the sequence is below a predetermined threshold S_pl2, and if the indicator R_3d determined for the current image or for a previous image in the sequence is below a predetermined threshold S_3d2,

reduced if the indicator R_pl determined for a previous image in the sequence is above a predetermined threshold S_pl2, and

unchanged otherwise.

10. The method as claimed in claim 9, wherein the reduction and increase of the width of the validity range are limited.

11. The method as claimed in claim 2, wherein the threshold d_sel is dependent on the criterion R_pl determined for a plurality of previous images in the sequence, the set of valid points of the current depth image being selected as points of interest if the criterion R_pl determined for a plurality of previous images in the sequence is below a predetermined threshold S_sel.

12. The method as claimed in claim 2, comprising a supplementary step of determining an indicator Q_p, representing the quality of the ground plane determined for the current depth image, said indicator Q_p being determined as a function of the indicators R_3d and R_pl determined for the current depth image.

13. The method as claimed in claim 12, wherein the indicator Q_p is determined by the relation:

14. The method as claimed in claim 2, wherein the indicator R_pl of a depth image is equal to the number of points in said image whose distance from the ground plane defined by the triplet of spatially filtered parameters is less than the predetermined distance d_p, divided by the number of valid points in this image.

15. The method as claimed in claim 1, wherein the indicator R_3d of a depth image is equal to the number of valid points in this image divided by the total number of points in this image.

16. A system (10) for determining a ground plane in a sequence of images, comprising at least a module configured to execute the method according to claim 1.

17. A computer program comprising instructions code for executing the method according to claim 1, when the program is executed by a processor.

18. A processor-readable recording medium on which is recorded a program including instructions for executing the method according to claim 1, when the program is executed by a processor.