Method of detecting structural parts of a scene
View Patent ↗A method of detecting the structural elements within a scene sensed by at least one sensor within a locale, the method comprising: a) capturing data from the sensor, which data provides a first representation of the sensed scene at the current time; b) generating a second representation of the sensed scene where the second representation is generated from a prior model of the locale; and c) comparing the first and second representations with one another to determine which parts of the first representation represent structural elements of the locale.
1. A method of detecting structural elements within a scene sensed by a first sensor within a locale, the method comprising:
acquiring data from the first sensor, which data provides a first representation of the sensed scene at a current time;
generating a second representation of the sensed scene where the second representation is generated from a prior model of the locale, the prior model of the locale generated using a second sensor different from the first sensor; and
comparing the first and second representations with one another to determine which parts of the first representation represent structural elements of the locale, wherein comparing the first and second representations comprises generating disparity differences between the first and second representations which are weighted according to an associated measurement uncertainty.
2. A method according to claim 1 in which generating the second representation includes generating the second representation by reprojecting at least a portion of the prior model to represent the first representation.
3. A method according to claim 2 in which the prior model is processed so that the second representation includes only the structural elements and excludes ephemeral objects.
4. A method according to claim 1 in which the measurement uncertainty is provided by using a pre-computed filter.
5. A method according to claim 4 in which the pre-computed filter is an average depth-Jacobian image where ∂z s /∂x is computed for each pixel in the following:
Z
x
:=
(
∂
z
s
∂
x
)
P
x
(
∂
z
s
∂
x
)
T
.
6. A method according to claim 5 in which the average depth Jacobian image is computed over a plurality of frames and the results combined.
7. A method according to claim 1 in which generating the second representation includes generating the second representation using a determination of movement between a time the first representation was provided and the sensed scene was provided by the prior model.
8. A method according to claim 1 in which the method is used to localize a vehicle within a vehicle locale, the first sensor is mounted upon the vehicle, and wherein the method is applied to data obtained from the first sensor on the vehicle.
9. A method according to claim 1 wherein the first sensor and the second sensor are of different sensor modalities.
10. A method according to claim 1 wherein the first sensor is a camera and the second sensor is a laser scanner.
11. A system comprising a memory, processing circuitry, and a first sensor, the system arranged to detect structural elements within a sensed scene, wherein the system is arranged to perform the following tasks:
acquire, using the first sensor, a first representation of a sensed scene taken at a first time;
generate, using the processing circuitry, a second representation of the sensed scene, where the second representation is generated from a prior model of a locale containing the sensed scene, the prior model of the locale having been generated using a second sensor different from the first sensor;
compare, using the processing circuitry, the first and second representations with one another to determine which parts of the first representation are used to determine structural elements of the locale; and
generate, using the processing circuitry, disparity differences between the first and second representations which are weighted according to an associated measurement uncertainty.
12. A system according to claim 11 which is arranged to reproject at least a portion of the prior model to provide a similar view to that of the first representation.
13. A system according to claim 11 which is arranged to generate the second representation using a determination of movement between the first time that the first representation was acquired and a different time.
14. A system according to claim 11 which is arranged to weight the first and second representations by a pre-computed filter.
15. A system according to claim 11 wherein the first sensor and the second sensor are of different sensor modalities.
16. A system according to claim 11 wherein the first sensor is a camera and the second sensor is a laser scanner.
17. A non-transitory machine readable medium containing instructions which when read by a computer cause the computer to:
acquire, using a first sensor, a first representation of a scene taken at a first time;
generate a second representation of the scene, where the second representation is generated from a prior model of a locale containing the scene, the prior model of the locale generated using a second sensor different from the first sensor;
compare the first and second representations with one another to determine which parts of the first representation are used to determine structural elements of the locale; and
generate disparity differences between the first and second representations which are weighted according to an associated measurement uncertainty.
18. A non-transitory machine readable medium according to claim 17 in which the measurement uncertainty is provided by using a pre-computed filter.
19. A non-transitory machine readable medium according to claim 18 in which the pre-computed filter is an average depth-Jacobian image where ∂z s /∂x is computed for each pixel in the following:
Z
x
:=
(
∂
z
s
∂
x
)
P
x
(
∂
z
s
∂
x
)
T
.
20. A non-transitory machine readable medium according to claim 19 in which the average depth Jacobian image is computed over a plurality of frames and the results combined.