Machine learning based target localization for autonomous unmanned vehicles

1. A system, comprising:

at least one processor; and

at least one memory including program code which when executed by the at least one processor provides operations comprising:

applying, to a first image captured by an apparatus, a first machine learning model trained to determine whether a target object is visible in one or more portions of the first image, the first image depicting a first region of a search area, and the one or more portions of the first image corresponding to a first plurality of sub-regions within the first region of the search area;

in response to the target object being absent from the first image of the first region of the search area, applying a second machine learning model trained to identify, based at least on an output of the first machine learning model indicating the target object as being visible and/or nonvisible in the one or more portions of the first image, a second region of the search area corresponding to one of the first plurality of sub-regions within the first region of the search area having a highest probability of including the target object; and

sending, to the apparatus, a command navigating the apparatus to the second region of the search area in order to capture a second image of the second region of the search area.

2. The system of claim 1 , wherein the first image of the first region is captured at a first distance relative to the search area, and wherein the apparatus is navigated, based at least on a second distance relative to the search area being associated with a higher visibility level than the first distance, to the second distance in order to capture the second image of the second region.

3. The system of claim 2 , wherein the second machine learning model is further trained to determine, based at least on a visibility of the target object determined by the second distance relative to the search area and by one or more environmental conditions, a first sensing mode for processing a second image of the second region of the search area.

4. The system of claim 3 , wherein the second machine learning model identifies the first sensing model in response to the second distance being associated with a maximum visibility level and the first machine learning model being unable to detect the target object in the second image by applying a second sensing mode.

5. The system of claim 1 , wherein the second machine learning model is configured to update, based at least on the output of the first machine learning model indicating the target object as being visible and/or nonvisible in the one or more portions of the first image, a probability distribution of the target object being present in one or more regions of the search area, and wherein the second region of the search area is identified based at least on the updated probability distribution.

6. The system of claim 1 , further comprising:

applying, to the second image of the second region of the search area, the first machine learning model in order to determine whether the target object is visible in one or more portions of the second image; and

in response to the target object being absent from the second image of the second region of the search area, navigating the apparatus to a third region of the search area having a next highest probability of including the target object.

7. The system of claim 6 , wherein the second region of the search area and the third region of the search area form an optimal search path spanning a three-dimensional space, and wherein the second machine learning model is trained to determine the optimal search path by at least maximizing a reward function corresponding to a movement cost of the apparatus moving to different regions along the optimal search path and/or a sensing cost of detecting a presence of the target object at each region of the search area along the optimal search path.

8. The system of claim 7 , wherein the second machine learning model is trained by applying one or more reinforcement learning techniques.

9. The system of claim 6 , wherein the third region of the search area comprises another one of the first plurality of sub-regions within the first region of the search area, and wherein the apparatus is navigated to the third region of the search area by at least navigating the apparatus in a same horizontal plane.

10. The system of claim 6 , wherein the third region of the search area comprises one of a second plurality of sub-regions within the second region of the search area, and wherein the apparatus is navigated to the third region of the search area by at least navigating the apparatus vertically to a different distance relative to the search area.

11. The system of claim 1 , wherein the apparatus comprises an autonomous unmanned vehicle.

12. The system of claim 1 , wherein the first machine learning model comprises a neural network.

13. The system of claim 1 , further comprising:

training, based at least on one or more training images, the first machine learning model, the one or more training images including a first image in which more than a threshold portion of the target object is visible and/or a second image in which less than the threshold portion of the target object is visible.

14. The system of claim 1 , wherein the second machine learning model comprises a Markov decision process.

15. A computer-implemented method, comprising:

applying, to a first image captured by an apparatus, a first machine learning model trained to determine whether a target object is visible in one or more portions of the first image, the first image depicting a first region of a search area, and the one or more portions of the first image corresponding to a first plurality of sub-regions within the first region of the search area;

in response to the target object being absent from the first image of the first region of the search area, applying a second machine learning model trained to identify, based at least on an output of the first machine learning model indicating the target object as being visible and/or nonvisible in the one or more portions of the first image, a second region of the search area corresponding to one of the first plurality of sub-regions within the first region of the search area having a highest probability of including the target object; and

sending, to the apparatus, a command navigating the apparatus to the second region of the search area in order to capture a second image of the second region of the search area.

16. The method of claim 15 , wherein the first image of the first region is captured at a first distance relative to the search area, and wherein the apparatus is navigated, based at least on a second distance relative to the search area being associated with a higher visibility level than the first distance, to the second distance in order to capture the second image of the second region.

17. The method of claim 16 , wherein the second machine learning model is further trained to determine, based at least on a visibility of the target object determined by the second distance relative to the search area and by one or more environmental conditions, a first sensing mode for processing a second image of the second region of the search area.

18. The method of claim 17 , wherein the second machine learning model identifies the first sensing model in response to the second distance being associated with a maximum visibility level and the first machine learning model being unable to detect the target object in the second image by applying a second sensing mode.

19. The method of claim 15 , wherein the second machine learning model is configured to update, based at least on the output of the first machine learning model indicating the target object as being visible and/or nonvisible in the one or more portions of the first image, a probability distribution of the target object being present in one or more regions of the search area, and wherein the second region of the search area is identified based at least on the updated probability distribution.

20. A non-transitory computer-readable medium storing instructions, which when executed by at least one data processor, result in operations comprising:

applying, to a first image captured by an apparatus, a first machine learning model trained to determine whether a target object is visible in one or more portions of the first image, the first image depicting a first region of a search area, and the one or more portions of the first image corresponding to a first plurality of sub-regions within the first region of the search area;

in response to the target object being absent from the first image of the first region of the search area, applying a second machine learning model trained to identify, based at least on an output of the first machine learning model indicating the target object as being visible and/or nonvisible in the one or more portions of the first image, a second region of the search area corresponding to one of the first plurality of sub-regions within the first region of the search area having a highest probability of including the target object; and

sending, to the apparatus, a command navigating the apparatus to the second region of the search area in order to capture a second image of the second region of the search area.