Neural network
A computer that includes a processor and a memory, the memory including instructions executable by the processor to plot three-dimensional positions of labeled cuboids from successive frames between a first keyframe and a second keyframe of a densely labeled video sequence. A sparsely labeled video sequence can be generated by removing labeled cuboids from the densely labeled video sequence with positions that are within a user-selected distance from a line extending from a position of the labeled cuboid in the first keyframe to a position of the labeled cuboid in the second keyframe. The sparsely labeled video sequence can be output for training a machine learning system.
1 . A system, comprising:
a computer that includes a processor and a memory, the memory including instructions executable by the processor to:
plot three-dimensional positions of labeled cuboids from successive frames between a first keyframe and a second keyframe of a densely labeled video sequence;
generate a sparsely labeled video sequence by removing the labeled cuboids from the densely labeled video sequence with positions that are within a user-selected distance from a three-dimensional line extending from a position of the labeled cuboid in the first keyframe to a position of the labeled cuboid in the second keyframe; and
output the sparsely labeled video sequence for training a machine learning system.
2 . The system of claim 1 , the instructions including further instruction to:
plot orientations of the labeled cuboids from the successive frames between the first keyframe and the second keyframe of the densely labeled video sequence; and
generate the sparsely labeled video sequence by removing the labeled cuboids from the densely labeled video sequence with orientations that are within the user-selected distance from a line extending from the orientation of the labeled cuboid in the first keyframe to the orientation of the labeled cuboid in the second keyframe.
3 . The system of claim 1 , wherein the labeled cuboids are three-sided wire frames that indicate objects.
4 . The system of claim 3 , wherein the labeled cuboids include labels that identify the objects.
5 . The system of claim 1 , wherein the three-dimensional positions of the labeled cuboids are x, y, and z global coordinates.
6 . The system of claim 1 , the instructions including further instructions to remove the labeled cuboids from the densely labeled video sequence by a Douglas-Peucker algorithm that removes the labeled cuboids by determining L2 distances between a three-dimensional polyline based on the positions of the labeled cuboids and the three-dimensional line extending from the position of the labeled cuboid in the first keyframe to the position of the labeled cuboid in the second keyframe.
7 . The system of claim 6 , wherein, when the L2 distance exceeds a maximum allowable L2 distance, a new keyframe is indicated in the video sequence.
8 . The system of claim 2 , the instructions including further instructions to remove the labeled cuboids from the densely labeled video sequence by a Douglas-Peucker algorithm that removes the labeled cuboids by determining distances by linear interpolation in quaternion space between a polyline based on the orientations of the labeled cuboids and the line extending from the orientation of the labeled cuboid in the first keyframe to the orientation of the labeled cuboid in the second keyframe.
9 . The system of claim 8 , wherein, when the distance exceeds a maximum allowable distance, a new keyframe is indicated in the video sequence.
10 . The system of claim 8 , wherein the Douglas-Peucker algorithm removes the labeled cuboids by determining distances between the polyline and the line extending from the orientation of the labeled cuboid in the first keyframe to the orientation of the labeled cuboid in the second keyframe based on L2 distances between Lie Algebra vectors for the line and the orientations of the labeled cuboids.
11 . The system of claim 8 , wherein the Douglas-Peucker algorithm removes the labeled cuboids by determining L2 distances in rotation matrices between the line and the orientations of the labeled cuboids.
12 . The system of claim 1 , wherein the machine learning system is a neural network.
13 . A method, comprising:
plotting three-dimensional positions of labeled cuboids from successive frames between a first keyframe and a second keyframe of a densely labeled video sequence;
generating a sparsely labeled video sequence by removing the labeled cuboids from the densely labeled video sequence with positions that are within a user-selected distance from a three-dimensional line extending from a position of the labeled cuboid in the first keyframe to a position of the labeled cuboid in the second keyframe; and
outputting the sparsely labeled video sequence for training a machine learning system.
14 . The method of claim 13 , further comprising:
plotting orientations of the labeled cuboids from the successive frames between the first keyframe and the second keyframe of the densely labeled video sequence; and
generating the sparsely labeled video sequence by removing labeled cuboids from the densely labeled video sequence with orientations that are within the user-selected distance from a line extending from the orientation of the labeled cuboid in the first keyframe to the orientation of the labeled cuboid in the second keyframe.
15 . The method of claim 13 , wherein the labeled cuboids are three-sided wire frames that indicate objects.
16 . The method of claim 15 , wherein the labeled cuboids include labels that identify the objects.
17 . The method of claim 13 , wherein the three-dimensional positions of the labeled cuboids are x, y, and z global coordinates.
18 . The method of claim 13 , further comprising removing the labeled cuboids from the densely labeled video sequence by a Douglas-Peucker algorithm that removes the labeled cuboids by determining L2 distances between a three-dimensional polyline based on the positions of the labeled cuboids and the three-dimensional line extending from the position of the labeled cuboid in the first keyframe to the position of the labeled cuboid in the second keyframe.
19 . The method of claim 18 , wherein, when the L2 distance exceeds a maximum allowable L2 distance, a new keyframe is indicated in the video sequence.
20 . The method of claim 14 , further comprising removing the labeled cuboids from the densely labeled video sequence by a Douglas-Peucker algorithm that removes the labeled cuboids by determining a distances by linear interpolation in quaternion space between a polyline based on the orientations of the labeled cuboids and the line extending from the orientation of the labeled cuboid in the first keyframe to the orientation of the labeled cuboid in the second keyframe.