Vehicle control apparatus and method thereof
View Patent ↗An apparatus for controlling vehicle driving comprises a sensor to obtain a cluster of points, a memory storing a neural network model, and a processor. The processor is configured to determine a first heading direction of a virtual box, generated from the cluster of points, by inputting the cluster into the neural network model. Based on identifying the first heading direction in one of several regions, the apparatus determines an angle between a reference axis within that region and the first heading direction. The regions are divided based on the virtual box's center. Based on this angle, the apparatus outputs the virtual box assigned with the first heading direction or a second heading direction opposite to the first heading direction. The apparatus outputs a signal associated with the virtual box's heading direction and controls vehicle driving based on this signal.
1 . An apparatus for controlling driving of a vehicle, the apparatus comprising:
a sensor configured to obtain a cluster of points;
a memory storing a neural network model; and
a processor configured to:
determine, based on inputting the cluster of points to the neural network model, a first heading direction of a virtual box, wherein the virtual box is generated based on the cluster of points;
determine, based on identifying the first heading direction in one region among a designated number of regions, an angle between a reference axis and the first heading direction, wherein the reference axis is included in the one region, and wherein the designated number of regions are divided based on a center of the virtual box;
based on the determined angle:
output the virtual box assigned with the first heading direction, wherein the first heading direction is maintained, or
output the virtual box assigned with a second heading direction that is opposite to the first heading direction; and
output a signal associated with:
the virtual box assigned with the first heading direction, or
the virtual box assigned with the second heading direction; and
control, based on the signal, driving of the vehicle.
2 . The apparatus of claim 1 , wherein the processor is configured to:
determine, based on assigning an index to each of the designated number of regions, a region index of the one region.
3 . The apparatus of claim 1 , wherein the processor is configured to:
output, based on the determined angle being less than a threshold angle, the virtual box assigned with the first heading direction.
4 . The apparatus of claim 1 , wherein the processor is configured to:
output, based on the determined angle being greater than or equal to a threshold angle, the virtual box assigned with the second heading direction.
5 . The apparatus of claim 1 , wherein the processor is configured to:
output, based on the determined angle and a position of the virtual box, the virtual box assigned with the first heading direction or the virtual box assigned with the second heading direction.
6 . The apparatus of claim 1 , wherein the processor is configured to:
train, based on a ground-truth (GT) bounding box and the cluster of points, the neural network model.
7 . The apparatus of claim 6 , wherein the processor is configured to:
train, based on information associated with the virtual box, the neural network model, and
wherein the information associated the virtual box comprises at least one of:
a position of the virtual box,
a size of the virtual box,
the first heading direction,
class information of the virtual box, or
score information of the virtual box.
8 . The apparatus of claim 7 , wherein the processor is configured to:
set a number of anchor boxes among result values output from the neural network model to a designated number to train the neural network model.
9 . The apparatus of claim 7 , wherein the processor is configured to:
obtain, based on training the neural network model, a loss value for at least one of the class information or the position.
10 . The apparatus of claim 6 , wherein the processor is configured to:
sequentially execute, based on performing pre-processing for the cluster of points, a first algorithm and a second algorithm; and
train, based on a result value obtained by sequentially executing the first algorithm and the second algorithm, the neural network model.
11 . A method performed by an apparatus for controlling driving of a vehicle, the method comprising:
determining, based on inputting a cluster of points to a neural network model, a first heading direction of a virtual box, wherein the virtual box is generated based on the cluster of points, and wherein the cluster of points are obtained by a sensor;
determining, based on identifying the first heading direction in one region among a designated number of regions, an angle between a reference axis and the first heading direction, wherein the reference axis is included in the one region, and wherein the designated number of regions are divided based on a center of the virtual box;
based on the determined angle:
outputting the virtual box assigned with the first heading direction, wherein the first heading direction is maintained, or
outputting the virtual box assigned with a second heading direction that is opposite to the first heading direction;
outputting a signal associated with:
the virtual box assigned with the first heading direction, or
the virtual box assigned with the second heading direction; and
controlling, based on the signal, driving of the vehicle.
12 . The method of claim 11 , further comprising:
determining, based on assigning an index to each of the designated number of regions, a region index of the one region.
13 . The method of claim 11 , wherein the outputting the virtual box comprises outputting, based on the determined angle being less than a threshold angle, the virtual box assigned with the first heading direction.
14 . The method of claim 11 , wherein the outputting the virtual box comprises outputting, based on the determined angle being greater than or equal to a threshold angle, the virtual box assigned with the second heading direction.
15 . The method of claim 11 , wherein the outputting the virtual box comprises outputting, based on the determined angle and a position of the virtual box, the virtual box assigned with the first heading direction or the virtual box assigned with the second heading direction.
16 . The method of claim 11 , further comprising:
training, based on a ground-truth (GT) bounding box and the cluster of points, the neural network model.
17 . The method of claim 16 , further comprising:
training, based on information associated with the virtual box, the neural network model,
wherein the information associated the virtual box comprises at least one of:
a position of the virtual box,
a size of the virtual box,
the first heading direction,
class information of the virtual box, or
score information of the virtual box.
18 . The method of claim 17 , further comprising:
setting a number of anchor boxes among result values output from the neural network model to a designated number to train the neural network model.
19 . The method of claim 17 , further comprising:
obtaining, based on training the neural network model, a loss value for at least one of the class information or the position.
20 . The method of claim 16 , further comprising:
sequentially executing, based on performing pre-processing for the cluster of points, a first algorithm and a second algorithm; and
training, based on a result value obtained by sequentially executing the first algorithm and the second algorithm, the neural network model.