Systems and methods for a real-time intelligent inspection assistant
Example implementations described herein are directed to a solution to the problem of accurate real-time inventory counting and industrial inspection. The solution, involves a device such as a mobile device that assists a human operator on the field to quickly achieve high quality inspection results. The example implementations detect objects of interest in individual image snapshots and use location and orientation sensors to integrate the snapshots to reconstruct a more accurate virtual representation of the inspection area. This representation can then be reorganized in various ways to derive inventory counts and other information that are not planned originally.
1. A method, comprising:
receiving, from a device comprising a camera, a global positioning satellite (GPS) sensor, and an inertial measurement unit (IMU) sensor; a plurality of image frames, each of the plurality of image frames comprising a subset of a plurality of objects;
receiving location data from the GPS sensor and orientation data from the IMU sensor associated with the each of the plurality of image frames;
identifying each of the subset of the plurality of objects for each of the plurality of image frames from utilization of bounding boxes;
determining translations between the plurality of image frames based on identified objects of the subset of the objects that are identical from overlapping portions of the plurality of image frames and the bounding boxes; and
determining the plurality of objects for the translations from overlapping portions of image frames to link continuous images with each other, wherein the translations are determined based on the location and orientation of the device for the each of the plurality of image frames as determined from the location data and the orientation data.
2. The method of claim 1 , wherein the translations are determined based on a similarity and a size of each of the subset of the plurality of objects in the plurality of image frames and on projection estimates of objects in different frames based on sensor data.
3. The method of claim 1 , further comprising receiving location data and orientation data of a device associated with the each of the plurality of image frames and wherein the translations are determined based on the location and orientation of the device for the each of the plurality of image frames.
4. The method of claim 3 , further comprising providing a virtual map indicating the plurality of objects and indicating a location and orientation of the device based on the location data and the orientation data.
5. The method of claim 1 , further comprising receiving depth data associated with each of the plurality of image frames, and determining a 3D reconstruction of the plurality of objects based on the depth data.
6. The method of claim 1 , wherein the identifying each of the subset of the plurality of objects for each of the plurality of image frames is conducted in real time.
7. The method of claim 1 , wherein the identifying the each of the subset of the plurality of objects for each of the plurality of image frames is conducted from a selection of a machine learning model configured to identify the plurality of objects.
8. The method of claim 7 , further comprising providing an interface configured to provide annotations to identify missing objects from the plurality of objects, and training the selected machine learning model from the annotations.
9. A device, comprising:
a camera;
a global positioning satellite (GPS) sensor;
an inertial measurement unit (IMU) sensor; and
a processor, configured to:
receive a plurality of image frames from the camera, each of the plurality of image frames
comprising a subset of a plurality of objects;
receive location data from the GPS sensor and orientation data from the IMU sensor associated with the each of the plurality of image frames;
identifying each of the subset of the plurality of objects for each of the plurality of image frames from utilization of bounding boxes;
determine translations between the plurality of image frames based on identified objects of the subset of the objects that are identical from overlapping portions of the plurality of image frames and the bounding boxes; and
determine the plurality of objects for the translations from overlapping portions of image frames to link continuous images with each other, wherein the translations are determined based on the location and orientation of the device for the each of the plurality of image frames as determined from the location data and the orientation data.
10. The device of claim 9 , wherein the processor is configured to determine the translations based on a similarity and a size of each of the subset of the plurality of objects in the plurality of image frames, and on projection estimates of objects in different frames based on sensor data.
11. The device of claim 9 , the processor configured to provide a virtual map indicating the plurality of objects and indicating a location and orientation of the device based on the location data and the orientation data.
12. The device of claim 9 , further comprising:
a Light Detection and Ranging (LiDAR) sensor;
wherein the processor is configured to receive depth data associated with each of the plurality of image frames from the LiDAR sensor, and determine a 3D reconstruction of the plurality of objects based on the depth data.
13. The device of claim 9 , wherein the processor is configured to identify each of the subset of the plurality of objects for each of the plurality of image frames in real time.
14. The device of claim 9 , wherein the processor is configured to identify the each of the subset of the plurality of objects for each of the plurality of image frames from a selection of a machine learning model configured to identify the plurality of objects.
15. The device of claim 14 , wherein the processor is further configured to provide an interface configured to provide annotations to identify missing objects from the plurality of objects, and train the selected machine learning model from the annotations.
16. A non-transitory computer readable medium, storing instructions for executing a process, the instructions comprising:
receiving, from a device comprising a camera, a global positioning satellite (GPS) sensor, and an inertial measurement unit (IMU) sensor; a plurality of image frames, each of the plurality of image frames comprising a subset of a plurality of objects;
receiving location data from the GPS sensor and orientation data from the IMU sensor associated with the each of the plurality of image frames;
identifying each of the subset of the plurality of objects for each of the plurality of image frames from utilization of bounding boxes;
determining translations between the plurality of image frames based on identified objects of the subset of the objects that are identical from overlapping portions of the plurality of image frames and the bounding boxes; and
determine the plurality of objects for the translations from overlapping portions of image frames to link continuous images with each other, wherein the translations are determined based on the location and orientation of the device for the each of the plurality of image frames as determined from the location data and the orientation data.