Systems and methods for detecting waste receptacles using convolutional neural networks

1. A system for detecting a waste receptacle, comprising:

a) a camera for capturing an image;

b) a convolutional neural network trained for identifying target waste receptacles, the convolutional neural network comprises a plurality of depthwise separable convolution filters and one of:

i) a MobileNet architecture, or

ii) a meta-architecture for object classification and bounding box regression; and

c) a processor mounted on the waste-collection vehicle, in communication with the camera and the convolutional neural network;

d) wherein the processor is configured for:

i) using the convolutional neural network to generate an object candidate based on the image;

ii) using the convolutional neural network to determine whether the object candidate corresponds to a target waste receptacle; and

iii) selecting an action based on whether the object candidate is acceptable.

2. The system of claim 1 , wherein the object candidate comprises an object classification and bounding box definition.

3. The system of claim 2 , wherein the use of the convolutional neural network to determine whether the object candidate corresponds to a target waste receptacle comprises:

a) predicting a class confidence score; and

b) if the class confidence score is greater than a pre-defined confidence threshold of acceptability, determining that the object candidate is acceptable;

c) otherwise determining that the object candidate is not acceptable.

4. The system of claim 2 , wherein the object classification comprises at least one of garbage, recycling, compost, and background.

5. The system of claim 2 , wherein the bounding box definition comprises pixel coordinates, a bounding box width, and a bounding box height.

6. The system of claim 1 , wherein the meta-architecture comprises single shot detection.

7. The system of claim 6 , wherein the meta-architecture comprises four additional convolution layers.

8. The system of claim 1 , further comprising:

a) an arm for grasping the waste receptacle, the arm being mountable on a waste- collection vehicle; and

b) an arm-actuation module connected to the arm, the arm-actuation module in communication with the processor;

c) wherein:

i) the processor is further configured for:

A. selecting the action of picking up the waste receptacle if the object candidate is acceptable;

B. selecting the action of rejecting the object candidate if the object candidate is not acceptable; and

C. if the action of picking up the waste receptacle is selected, calculating a location of the waste receptacle; and

ii) the arm-actuation module is configured for automatically moving the arm in response to the location of the waste receptacle.

9. The system of claim 8 , wherein the arm-actuation module is configured so that the moving the arm comprises grasping the waste receptacle.

10. The system of claim 9 , wherein the moving the arm further comprises lifting the waste receptacle and dumping contents of the waste receptacle into the waste-collection vehicle.

11. A method for detecting a waste receptacle comprising:

a) capturing an image with a camera;

b) using a convolutional neural network to generate an object candidate based on the image, wherein the convolutional neural network comprises a plurality of depthwise separable convolution filters and one of:

i) a MobileNet architecture, or

ii) a meta-architecture for object classification and bounding box regression;

c) determining whether the object candidate corresponds to a target waste receptacle; and

d) selecting an action based on whether the object candidate is acceptable.

12. The method of claim 11 , wherein the object candidate comprises an object classification and bounding box definition.

13. The method of claim 12 , wherein the using the convolutional neural network to determine whether the object candidate corresponds to a target waste receptacle comprises:

a) predicting a class confidence score; and

b) if the class confidence score is greater than a pre-defined confidence threshold of acceptability, determining that the object candidate is acceptable;

c) otherwise determining that the object candidate is not acceptable.

14. The method of claim 12 , wherein the object classification comprises at least one of garbage, recycling, compost, and background.

15. The method of claim 12 , wherein the bounding box definition comprises pixel coordinates, a bounding box width, and a bounding box height.

16. The method of claim 11 , wherein the meta-architecture comprises single shot detection.

17. The method of claim 16 , wherein the meta-architecture comprises four additional convolution layers.

18. The method of claim 11 wherein:

a) the selecting an action based on whether the object candidate is acceptable comprises:

i) selecting the action of picking up the waste receptacle if the object candidate is acceptable; and

ii) selecting the action of rejecting the object candidate if the object candidate is not acceptable; and

b) the method further comprises if and only if the action of picking up the waste receptacle is selected:

i) calculating a location of the waste receptacle; and

ii) moving an arm mounted on a waste-collection vehicle in response to the location of the waste receptacle.

19. The method of claim 18 , wherein the moving the arm comprises grasping the waste receptacle.

20. The method of claim 19 , wherein the moving the arm further comprises lifting the waste receptacle and dumping contents of the waste receptacle into the waste-collection vehicle.