Enhanced static object classification using LiDAR
Devices, systems, and methods are provided for classifying detected objects as static or dynamic. A device may determine first light detection and ranging (LIDAR) data associated with a convex hull of an object at a first time, and determine second LIDAR data associated with the convex hull at a second time after the first time. The device may generate, based on the first LIDAR data and the second LIDAR data, a vector including values of features associated with the first convex hull and the second convex hull. The device may determine, based on the vector, a probability that the object is static. The device may operate a machine based on the probability that the object is static.
1 . A method for classifying objects as static or dynamic, the method comprising:
generating, based on first LIDAR data associated with a convex hull of an object at a first time and second LIDAR data associated with the convex hull at a second time after the first time, a vector comprising values of features associated with the convex hull;
comparing a probability associated with the object to a threshold level to determine whether the object is static or dynamic, wherein the probability is based on the vector; and
adjusting a length of a buffer based on whether the probability indicates the object is static or dynamic by:
storing the second LIDAR data with previously detected LIDAR data in the buffer to increase the length of the buffer, wherein the previously detected LIDAR data includes the first LIDAR data, or
removing one or more of the previously detected LIDAR data until the length of the buffer is at a desired length to reduce the length of the buffer, wherein the second LIDAR data is stored in the buffer having the desired length.
2 . The method of claim 1 , wherein the probability is based on an expected value associated with the vector.
3 . The method of claim 1 , wherein the probability is based on a feature weight matrix.
4 . The method of claim 1 , wherein comparing the probability to the threshold level further comprises comparing the probability to a plurality of threshold levels.
5 . The method of claim 4 , wherein:
the probability satisfying a first threshold level of the plurality of threshold levels indicates a high confidence that the object is static;
the probability satisfying a second threshold level of the plurality of threshold levels indicates a low confidence the object is static; and
the probability failing to satisfy either the first threshold level or the second threshold level indicates that the object is dynamic.
6 . The method of claim 1 , wherein generating the vector is based on a transform between the first LIDAR data and the second LIDAR data.
7 . The method of claim 1 , wherein adjusting the length of the buffer comprises:
extending a previous length of the buffer in response to the probability indicating that the object is static; and
reducing the previous length of the buffer in response to the probability indicating that the object is dynamic.
8 . A system comprising:
a memory; and
a processor coupled to the memory and configured to:
generate, based on first LIDAR data associated with a convex hull of an object at a first time and second LIDAR data associated with the convex hull at a second time after the first time, a vector comprising values of features associated with the convex hull;
compare a probability associated with the object to a threshold level to determine whether the object is static or dynamic, wherein the probability is based on the vector; and
adjust a length of a buffer based on whether the probability indicates the object is static or dynamic by:
storing the second LIDAR data with previously detected LIDAR data in the buffer to increase the length of the buffer, wherein the previously detected LIDAR data includes the first LIDAR data, or
removing one or more of the previously detected LIDAR data until the length of the buffer is at a desired length to reduce the length of the buffer, wherein the second LIDAR data is stored in the buffer having the desired length.
9 . The system of claim 8 , wherein the probability is based on an expected value associated with the vector.
10 . The system of claim 8 , wherein the probability is based on a feature weight matrix.
11 . The system of claim 8 , wherein the processor configured to compare the probability to the threshold level is further configured to compare the probability to a plurality of threshold levels.
12 . The system of claim 11 , wherein:
the probability satisfying a first threshold level of the plurality of threshold levels indicates a high confidence that the object is static;
the probability satisfying a second threshold level of the plurality of threshold levels indicates a low confidence the object is static; and
the probability failing to satisfy either the first threshold level or the second threshold level indicates that the object is dynamic.
13 . The system of claim 8 , wherein the processor configured to generate the vector is further configured to generate the vector is based on a transform between the first LIDAR data and the second LIDAR data.
14 . The system of claim 8 , wherein the processor configured to adjust the length of the buffer is further configured to:
extend a previous length of the buffer in response to the probability indicating that the object is static; and
reduce the previous length of the buffer in response to the probability indicating that the object is dynamic.
15 . A non-transitory computer-readable medium having instructions stored thereon that, when executed by at least one computing device, causes the at least one computing device to perform operations, the operations comprising:
generating, based on first LIDAR data associated with a convex hull of an object at a first time and second LIDAR data associated with the convex hull at a second time after the first time, a vector comprising values of features associated with the convex hull;
comparing a probability associated with the object to a threshold level to determine whether the object is static or dynamic, wherein the probability is based on the vector; and
adjusting a length of a buffer based on whether the probability indicates the object is static or dynamic by:
storing the second LIDAR data with previously detected LIDAR data in the buffer to increase the length of the buffer, wherein the previously detected LIDAR data includes the first LIDAR data, or
removing one or more of the previously detected LIDAR data until the length of the buffer is at a desired length to reduce the length of the buffer, wherein the second LIDAR data is stored in the buffer having the desired length.
16 . The non-transitory computer-readable medium of claim 15 , wherein the probability is based on an expected value associated with the vector.
17 . The non-transitory computer-readable medium of claim 15 , wherein the probability is based on a feature weight matrix.
18 . The non-transitory computer-readable medium of claim 15 , wherein comparing the probability to the threshold level further comprises comparing the probability to a plurality of threshold levels.
19 . The non-transitory computer-readable medium of claim 18 , wherein:
the probability satisfying a first threshold level of the plurality of threshold levels indicates a high confidence that the object is static;
the probability satisfying a second threshold level of the plurality of threshold levels indicates a low confidence the object is static; and
the probability failing to satisfy either the first threshold level or the second threshold level indicates that the object is dynamic.
20 . The non-transitory computer-readable medium of claim 15 , wherein generating the vector is based on a transform between the first LIDAR data and the second LIDAR data, and wherein adjusting the length of the buffer comprises:
extending a previous length of the buffer in response to the probability indicating that the object is static; and
reducing the previous length of the buffer in response to the probability indicating that the object is dynamic.