Object recognition device and object recognition method

Patent metadata

Field	Value
Publication number	US-11055582-B2
Application number	US-201916681294-A
Country	US
Kind code	B2
Filing date	Nov 12, 2019
Priority date	Dec 10, 2018
Publication date	Jul 6, 2021
Grant date	Jul 6, 2021

Abstract

Provided is an object recognition device capable of preventing a change of fusion data from a previous value to a current value that exceeds a tolerable range. The object recognition device is configured to execute specialized tracking processing when a change of the fusion data from the previous value to the current value exceeds a tolerable range, and generate tracking data that is equivalent to prediction data in the specialized tracking processing by setting an adjustment physical quantity for solving a fusion data discontinuity state.

First claim

What is claimed is:

1. An object recognition device, which is configured to generate fusion data from observation data that is received from each of a plurality of sensors to output the fusion data, each of the plurality of sensors being configured to detect, as the observation data, a physical quantity that indicates a state of an object present in a detectable range, the object recognition device comprising:
a data receiver configured to receive the observation data from each of the plurality of sensors;
a prediction processor configured to generate prediction data that is a result of predicting a current value of the fusion data by using a previous value of the fusion data and in accordance with a specific movement model to output the prediction data; and
a tracking processor configured to execute specialized tracking processing when a change of the fusion data from the previous value to the current value exceeds a tolerable range,
the tracking processor being configured to generate, in the specialized tracking processing, tracking data that is equivalent to the prediction data by setting an adjustment physical quantity for solving a state in which a change of the fusion data from the previous value to the current value exceeds the tolerable range.

2. The object recognition device according to claim 1 , further comprising a fusion result output processor configured to output the current value of the fusion data,
wherein the fusion result output processor is configured to replace a predicted value of a specific physical quantity included in the prediction data that is equivalent to the tracking data output from the tracking processor with the adjustment physical quantity set by the tracking processor, and output the prediction data processed by the replacement as the current value of the fusion data.

3. The object recognition device according to claim 2 ,
wherein the specific movement model is a uniformly accelerated motion model, and
wherein each of the adjustment physical quantity and the specific physical quantity is acceleration.

4. The object recognition device according to claim 1 ,
wherein the tracking processor is configured to perform discontinuity determination for determining whether the state in which a change of the fusion data from the previous value to the current value exceeds the tolerable range is happening, and switch tracking processing to be executed between normal tracking processing and the specialized tracking processing based on a result of the discontinuity determination, and
wherein the tracking processor is configured to generate tracking data in the normal tracking processing through an update of the previous value of the tracking data by using the prediction data output from the prediction processor and the observation data associated with the prediction data to output the generated tracking data.

5. The object recognition device according to claim 4 , wherein the tracking processor is configured to determine in the discontinuity determination whether the state in which a change of the fusion data from the previous value to the current value exceeds the tolerable range is happening, based on a positional difference between a fusion data target value, which is calculated from the observation data received from each of the plurality of sensors, and the prediction data, which is output from the prediction processor.

6. The object recognition device according to claim 4 , wherein the tracking processor is configured to perform the discontinuity determination when a number of sensors associated with the fusion data is changed.

7. The object recognition device according to claim 4 , wherein the tracking processor is configured to continue the specialized tracking processing when it is determined as a result of the discontinuity determination after a switch from the normal tracking processing to the specialized tracking processing that the state in which a change of the fusion data from the previous value to the current value exceeds the tolerable range is happening.

8. The object recognition device according to claim 4 , wherein the tracking processor is configured to make a switch from the specialized tracking processing to the normal tracking processing when it is determined as a result of the discontinuity determination after a switch from the normal tracking processing to the specialized tracking processing that the state in which a change of the fusion data from the previous value to the current value exceeds the tolerable range is not happening.

9. The object recognition device according to claim 4 , wherein the tracking processor is configured to execute prevention processing for preventing the state in which a change of the fusion data from the previous value to the current value exceeds the tolerable range from occurring, when a switch from the specialized tracking processing to the normal tracking processing is made.

10. The object recognition device according to claim 9 , wherein, as the prevention processing, the tracking processor saves an error covariance matrix that has been used in the normal tracking processing when a switch from the normal tracking processing to the specialized tracking processing is made, and uses the saved error covariance matrix to execute the normal tracking processing when a switch from the specialized tracking processing to the normal tracking processing is made.

11. An object recognition method for generating fusion data from observation data that is received from each of a plurality of sensors and output the fusion data, each of the plurality of sensors being configured to detect, as the observation data, a physical quantity that indicates a state of an object present in a detectable range, the object recognition method comprising:
receiving the observation data from each of the plurality of sensors;
generating prediction data that is a result of predicting a current value of the fusion data by using a previous value of the fusion data and in accordance with a specific movement model to output the prediction data; and
executing specialized tracking processing when a change of the fusion data from the previous value to the current value exceeds a tolerable range,
the executing specialized tracking processing including generating, in the specialized tracking processing, tracking data that is equivalent to the prediction data by setting an adjustment physical quantity for solving a state in which a change of the fusion data from the previous value to the current value exceeds the tolerable range.

Assignees

• Mitsubishi Electric Corp
• Mazda Motor

Inventors

• Tateiwa Shinichi
• Iida Koji
• Moritani Takayuki

Classifications

• G06F18/251 Primary

  of input or preprocessed data · CPC title

• G06F18/25

  Fusion techniques · CPC title

• G06V2201/07

  Target detection · CPC title

• G06K9/6288 Primary

  Physics · mapped topic

• G06K2209/21

  Physics · mapped topic