Control system using autoencoder
A control system comprises a memory storing a sequence of sensor data received from one or more sensors. The control system has a processor which processes the sensor data to compute a sequence of derived sensor data values. An autoencoder receives the sequence of derived sensor data values and computes a forward prediction of the sequence of derived sensor data values, the autoencoder having been trained imposing a relationship on positions of the derived sensor data values encoded in a latent space of the autoencoder. A processor initiates control of an apparatus using the forward prediction.
1. A control system comprising:
a memory comprising a parameterized model of an object;
a processor; and
a pose tracker that causes the processor to perform the following operations:
access the parameterized model of an object;
receive captured data depicting the object;
apply current values of parameters to the parameterized model of the object;
compute a correspondence for each of a plurality of observed data points in the captured data;
compute an optimization over the current values of the parameters and the correspondences; and
based at last on computing the optimization, update the current values of the parameters and the correspondences, wherein a decision to update the current values of the parameters is based at least on one or more factors comprising a current performance of the control system, a confidence level of an output of an autoencoder, a difference in a number of parameters of a latent space of the autoencoder and the number of parameters of the parameterized model of the object, a difference in a type of parameters of the latent space of the autoencoder and the parameters of the parameterized model of the object.
2. The control system of claim 1 wherein the parameterized model of the object is a polygon mesh model of the object.
3. The control system of claim 1 wherein the captured data is one of a 3D point cloud, a depth map, one or more frames of raw time of flight data, color image data or other captured data depicting the object to be tracked.
4. The control system of claim 1 , wherein the pose tracker further causes the processor to extract a region of interest from the captured data where the region of interest depicts the object to be tracked.
5. The control system of claim 4 , wherein the region of interest comprises parts of a depth map, and the pose tracker further causes the processor to compute a 3D point cloud by back projecting the region of interest.
6. The control system of claim 1 wherein the pose tracker further causes the processor to carry out the updating to the correspondences by moving the correspondences along a surface of the parameterized model of the object by an amount and in a direction specified by the computed optimization.
7. The control system of claim 1 wherein the pose tracker further causes the processor to compute a 3D point cloud is captured by a scanning sensor.
8. A method comprising:
accessing a parameterized model of an object;
receiving captured data depicting the object;
applying current values of parameters to the parameterized model of the object;
computing a correspondence for each of a plurality of observed data points in the captured data;
computing an optimization over the current values of the parameters and the correspondences; and
based at last on computing the optimization, update the current values of the parameters and the correspondences, wherein a decision to update the current values of the parameters is based at least on one or more factors comprising a current performance of a control system, a confidence level of an output of an autoencoder, a difference in a number of parameters of a latent space of the autoencoder and the number of parameters of the parameterized model of the object, a difference in a type of parameters of the latent space of the autoencoder and the parameters of the parameterized model of the object.
9. The method of claim 8 wherein the parameterized model of the object is a polygon mesh model of the object.
10. The method of claim 8 wherein the captured data is one of a 3D point cloud, a depth map, one or more frames of raw time of flight data, color image data or other captured data depicting the object to be tracked.
11. The method of claim 8 , further comprising extracting a region of interest from the captured data where the region of interest depicts the object to be tracked.
12. The method of claim 11 , wherein the region of interest comprises parts of a depth map, and a pose tracker that further causes the processor to compute a 3D point cloud by back projecting the region of interest.
13. The method of claim 8 , further comprising carrying out the updating to the correspondences by moving the correspondences along a surface of the parameterized model of the object by an amount and in a direction specified by the computed optimization.
14. The method of claim 8 further comprising capturing a 3D point cloud by a scanning sensor.
15. One or more computer storage media comprising computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the following operations:
accessing a parameterized model of an object;
receiving captured data depicting the object;
applying current values of parameters to the parameterized model of the object;
computing a correspondence for each of a plurality of observed data points in the captured data;
computing an optimization over the current values of the parameters and the correspondences; and
based at last on computing the optimization, update the current values of the parameters and the correspondences, wherein a decision to update the current values of the parameters is based at least on one or more factors comprising a current performance of a control system, a confidence level of an output of an autoencoder, a difference in a number of parameters of a latent space of the autoencoder and the number of parameters of the parameterized model of the object, a difference in a type of parameters of the latent space of the autoencoder and the parameters of the parameterized model of the object.
16. The computer storage media of claim 15 wherein the parameterized model of the object is a polygon mesh model of the object.
17. The computer storage media of claim 15 , wherein the captured data is one of a 3D point cloud, a depth map, one or more frames of raw time of flight data, color image data or other captured data depicting the object to be tracked.
18. The computer storage media of claim 15 , wherein the computer-executable instructions further cause the comprising extracting a region of interest from the captured data where the region of interest depicts the object to be tracked.
19. The computer storage media of claim 18 , wherein the region of interest comprises parts of a depth map, and a pose tracker that further causes the processor to compute a 3D point cloud by back projecting the region of interest.
20. The computer storage media of claim 15 , further comprising carrying out the updating to the correspondences by moving the correspondences along a surface of the parameterized model of the object by an amount and in a direction specified by the computed optimization.