Training of machine learning sensor data classification system
This application discloses training of a classification system for an assisted or automated driving system of a vehicle. A processing system can label sensor measurement data collected by sensors mounted in the vehicle with classifications, which can include a type of an object associated with the sensor measurement data and a confidence level of the classification. A training system can utilize the classifications labeled to the sensor measurement data to train a classification graph utilized by the classification system. The training system can select a node in a classification graph based, at least in part, on a classification labeled to sensor measurement data. The training system can compare the sensor measurement data to matchable data in the selected node, and modify the classification graph based, at least in part, on differences between the sensor measurement data and the matchable data in the selected node.
1. A method comprising:
selecting, by a computing system, a node in a classification graph based, at least in part, on a classification labeled to sensor measurement data collected by sensors mounted in a vehicle;
comparing, by the computing system, the sensor measurement data to matchable data in the selected node to identify differences between the sensor measurement data and the matchable data in the selected node; and
modifying, by the computing system, the classification graph by generating a new node for the classification graph, wherein the new node includes matchable data based, at least in part, on the differences between the sensor measurement data and the matchable data in the selected node, wherein a control system is configured to control vehicle operations based, at least in part, on object classifications identified by the modified classification graph.
2. The method of claim 1 , wherein modifying the classification graph further comprising altering the matchable data in the selected node based, at least in part, on the differences between the sensor measurement data and the matchable data in the selected node.
3. The method of claim 1 , wherein generating the new node includes linking the new node to the selected node in the classification graph, wherein the link between the selected node and the new node allows for traversal between the selected node and the new node in the classification graph during classification operations.
4. The method of claim 1 , further comprising:
identifying a first set of the sensor measurement data is correlated to a second set of the sensor measurement data; and
re-labeling the first set of the sensor measurement data with a classification associated with the second set of the sensor measurement data.
5. The method of claim 4 , wherein re-labeling the first set of the sensor measurement data with the classification associated with the second set of the sensor measurement data based on a confidence level of the classification associated with the second set of the sensor measurement data.
6. The method of claim 1 , wherein the matchable data includes an object model having at least one of an object pose, an object orientation, a transitional state, an object deformation, or a textural feature.
7. An apparatus comprising at least one memory device storing instructions configured to cause one or more processing devices to perform operations comprising:
selecting a node in a classification graph based, at least in part, on a classification labeled to sensor measurement data collected by sensors mounted in a vehicle;
identifying a first set of the sensor measurement data is correlated to a second set of the sensor measurement data;
re-labeling the first set of the sensor measurement data with a classification associated with the second set of the sensor measurement data;
comparing the sensor measurement data to matchable data in the selected node to identify differences between the sensor measurement data and the matchable data in the selected node; and
modifying the classification graph based, at least in part, on the differences between the sensor measurement data and the matchable data in the selected node, wherein a control system is configured to control vehicle operations based, at least in part, on object classifications identified by the modified classification graph.
8. The apparatus of claim 7 , wherein modifying the classification graph further comprising altering the matchable data in the selected node based, at least in part, on the differences between the sensor measurement data and the matchable data in the selected node.
9. The apparatus of claim 7 , wherein modifying the classification graph further comprising generating a new node for the classification graph, wherein the new node includes matchable data based, at least in part, on the differences between the sensor measurement data and the matchable data in the selected node.
10. The apparatus of claim 9 , wherein generating the new node includes linking the new node to the selected node in the classification graph, wherein the link between the selected node and the new node allows for traversal between the selected node and the new node in the classification graph during classification operations.
11. The apparatus of claim 7 , wherein re-labeling the first set of the sensor measurement data with the classification associated with the second set of the sensor measurement data based on a confidence level of the classification associated with the second set of the sensor measurement data.
12. The apparatus of claim 7 , wherein the matchable data includes an object model having at least one of an object pose, an object orientation, a transitional state, an object deformation, or a textural feature.
13. A system comprising:
a memory device configured to store machine-readable instructions; and
a computing system including one or more processing devices, in response to executing the machine-readable instructions, configured to:
select a node in a classification graph based, at least in part, on a classification labeled to sensor measurement data collected by sensors mounted in a vehicle;
compare the sensor measurement data to matchable data in the selected node to identify differences between the sensor measurement data and the matchable data in the selected node; and
modify the classification graph by generating a new node for the classification graph, wherein the new node includes matchable data based, at least in part, on the differences between the sensor measurement data and the matchable data in the selected node, wherein a control system is configured to control vehicle operations based, at least in part, on object classifications identified by the modified classification graph.
14. The system of claim 13 , wherein the one or more processing devices, in response to executing the machine-readable instructions, are configured to modify the classification graph by altering the matchable data in the selected node based, at least in part, on the differences between the sensor measurement data and the matchable data in the selected node.
15. The system of claim 13 , wherein the one or more processing devices, in response to executing the machine-readable instructions, are configured to generate the new node by linking the new node to the selected node in the classification graph, wherein the link between the selected node and the new node allows for traversal between the selected node and the new node in the classification graph during classification operations.
16. The system of claim 13 , wherein the one or more processing devices, in response to executing the machine-readable instructions, are configured to:
identify a first set of the sensor measurement data is correlated to a second set of the sensor measurement data; and
re-label the first set of the sensor measurement data with a classification associated with the second set of the sensor measurement data.
17. The system of claim 16 , wherein the one or more processing devices, in response to executing the machine-readable instructions, are configured to re-label the first set of the sensor measurement data with the classification associated with the second set of the sensor measurement data based on a confidence level of the classification associated with the second set of the sensor measurement data.