VISUALLY CURED RANDOM ACCESS LIDAR SYSTEM AND METHOD FOR LOCATION AND NAVIGATION - Patent application

Abstract

A visually cued random access LIDAR system and method for determining the location and / or navigation path of a mobile platform. A vision system on the mobile platform identifies a region of interest. The system classifies objects within the region of interest and directs a random access LIDAR to ping one or more of the classified objects. The platform locates in three dimensions using data from the vision system and LIDAR. The classification, direction, and identification steps are preferably performed continuously while the platform is moving and / or while the field of view (FOV) of the vision system is changing. Objects are preferably classified using at least one smart vision algorithm, such as a machine learning algorithm.

Description

[Technical field]

[0001] (Government Rights) This invention was made with Government support under HR0011-16-C-0105 awarded by DARPA's MOABB (Modular Optical Aperture Building Blocks). The Government has certain rights in this invention.

[0002] The present invention relates generally to a system and method for determining the location of a mobile platform relative to selected reference features and planning a safe route for navigation. [Background technology]

[0003] The field of autonomous vehicles is growing rapidly. Such vehicles, whether ground or airborne, need technology to know where they are (localization) and where they are going (navigation).

[0004] Various technologies are used to provide the necessary information. For example, magnetometers can be used for location and navigation, but they tend to be less accurate and sensitive to electromagnetic fields. Other physical sensors, such as odometers, are large and inflexible. GPS systems are prone to jamming and do not work well indoors. Time-of-flight cameras can be employed, but they tend to be pixel limited and have poor lateral resolution. Traditional scanning LIDAR can be considered, but this approach is highly inefficient, generates unnecessarily large amounts of data, and is not compatible with small platforms with high platform obstructions. Summary of the Invention

[0005] A vision-cued random access LIDAR system and method is presented that addresses some of the above problems.

[0006] The method disclosed herein is used to determine the location and / or navigation path of a mobile platform. The method involves using a vision system on the mobile platform to identify an area of ​​interest, classifying objects within the area of ​​interest, pointing a random access LIDAR to ping one or more of the classified objects, and identifying the location of the platform in three dimensions using data from the vision system and the LIDAR. LIDAR pings may also be used to determine the orientation of classified objects within the area of ​​interest.

[0007] The steps of classifying, directing and identifying are preferably performed continuously while the platform is moving and / or the field-of-view (FOV) of the vision system is changing. Objects are preferably classified using at least one smart vision algorithm, such as a machine learning algorithm.

[0008] The LIDAR system and method may also be used to determine the linear and / or angular acceleration and / or linear and / or angular velocity of the platform. The LIDAR may be a coherent LIDAR that may be employed to determine the velocity of classified objects relative to the moving platform.

[0009] These and other features, aspects, and advantages of the present invention will become better understood with reference to the following drawings, description, and claims. [Brief description of the drawings]

[0010] [Figure 1] 1 is a flow diagram illustrating one possible version of the methods described herein.

[0011] [Figure 2a]FIG. 1 is a block diagram illustrating one possible embodiment of the system described herein. [Figure 2b] FIG. 1 is a block diagram illustrating one possible embodiment of the system described herein. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] The system and method determine the location and / or navigation path of a moving platform. The invention has many possible applications. For example, the system and method can be used as part of an autonomously navigated ground or air vehicle. Such vehicles need a means to follow a specified path and identify and possibly avoid obstacles, such as power lines, along said path. The use of a vision system in combination with the random access LIDAR described herein allows the vehicle to travel at speeds of 50 mph or more. Another possible application is to provide location and / or navigation information in GPS-denied environments. A random access LIDAR is defined as a LIDAR that is capable of pointing its beam from any point within its addressable FOV to any other point within its addressable FOV very quickly (<1 ms), preferably without limitations imposed by mechanical inertia (see, for example, Poulton et al., Optics Letters Vol. 42, No. 20 / October 15, 2017).

[0013] Briefly, the system and method uses a vision system to queue random access LIDAR to determine the location of a moving platform relative to selected reference features. The reference features are preferably continuously identified and updated, preferably using smart vision algorithms, as the platform moves and / or the FOV changes. A simplified diagram of the method is shown in FIG. 1. The method includes: using a vision system on the moving platform to identify an area of ​​interest (step 10); Classifying (12) objects within the region of interest; directing the random access LIDAR to ping one or more of the classified objects (14); Locating the platform in three dimensions using data from the vision system and LIDAR (16); Includes. A system according to the invention (shown in FIG. 2) includes a software-based processor (36) that receives data from the vision system and classifies objects within the region of interest. LIDAR may optionally be used to determine the orientation of classified objects within the region of interest.

[0014] The steps of classifying objects within the region of interest, directing the random access LIDAR to ping one or more of the classified objects, and locating the platform in three dimensions using data from the vision system and the LIDAR are preferably performed continuously while the platform is moving and / or while the FOV of the vision system is changing. One way this can be done is by operating the vision system so that it captures images of the region of interest with a known interval between image captures, and then defining (18) at least three non-collinear static points within the region of interest to be used as common reference points for each image capture. The interval between image captures is preferably such that subsequent images have enough common features to allow at least three non-collinear features to be selected as common reference points. New reference features are added as old reference points move out of the FOV. It is noted that if more than two non-collinear features are available, redundancy can improve system accuracy. Additionally, it should be noted that the vision system and the random access LIDAR may be calibrated to establish a predetermined relationship between their respective FOVs.

[0015] Objects within the region of interest may be classified using at least one smart vision algorithm, such as, for example, a machine learning algorithm, where, for example, at least three non-collinear static points may be defined within the region of interest to be used as a common reference point for each image capture, and classification of the objects via the smart vision algorithm is used to select the static reference points.

[0016] The LIDAR preferably provides a narrow beam and is electro-optically pointed to give the required random access. The LIDAR is preferably a coherent LIDAR that can be used to determine (20) the velocity of objects in a region of interest relative to a moving platform. Coherent LIDAR can use the Doppler effect to measure the velocity of classified objects, thereby eliminating the need to compare two range measurements to obtain the longitudinal velocity. If the object is stationary (e.g. a traffic sign), the measurement is the velocity of the platform itself. Coherent LIDAR can also be configured to report the vibration signature of the object in addition to the amplitude and phase, which can be used by the processor software during the classification process.

[0017] The processor software may also be configured to use a plurality of points on the classified objects of interest to orient the objects to a correct pose. The classified objects and poses may be used to build a 3D model of the world, and the 3D model can be used to determine routes that are available for navigation to the next waypoint.

[0018] Previous systems employing scanning LIDAR tend to be overloaded with incoming data. The present combination of a vision system that identifies an area of ​​interest that queues a random-access LIDAR to ping classified objects within the area of ​​interest, and images of objects within the area, allows for 3D position data to be obtained in real time.

[0019] The random access LIDAR may be configured to perform multiple functions simultaneously based on a priority scheme established by the processor software. These functions include: scanning ahead to detect obstacles; a feedforward stabilization step to compensate for platform disturbances (described below); ranging points per cue received from the software interrogating objects of interest; may include:

[0020] There are at least two ways to determine that an object in the region of interest is stationary. One method is to measure the velocity of the object using, for example, a coherent LIDAR and compare the object velocity to a platform velocity measured using, for example, an inertial measurement unit (IMU). By this method, it can be determined whether the object is stationary. When so configured, the relative velocity of the object measured using the coherent LIDAR can be used to select at least three non-collinear static points to be used as a common reference point for each image capture in the region of interest.

[0021] The second method involves the use of smart vision algorithms for object classification as described above. If an object within a region of interest is classified as, for example, a tree or a traffic sign using a smart vision algorithm, the object may be considered to be static since some types of objects are static in nature.

[0022] The vision system required for the method may operate in the visible and / or thermal parts of the spectrum. Preferably, one or more cameras with high resolution may be used. The random access LIDAR preferably has a narrow instantaneous FOV (IFOV) that can be steered from any point in a wide field of regard to any other point when the FOV is for a short period of time. The IFOV is preferably 10μRad to 10mRad, with a field-of-regard of >10 degrees and a repointing time of 1μs to 1ms. The LIDAR preferably produces a diffraction limited beam.

[0023] The method for determining the location and / or navigation path of a mobile platform may further include a step (22) of determining the linear acceleration and / or angular acceleration and / or linear velocity and / or angular velocity of the platform. The required vision system preferably controls the pointing of the LIDAR used to ping one or more classified objects. When so configured, the linear acceleration and / or angular acceleration and / or linear velocity and / or angular velocity may be employed to apply corrections to the LIDAR to maintain a stable pointing directed by the vision system and latency between the vision system and the LIDAR when the platform is moving. Also, a predefined scan may be used to effectively increase the spot size on a detected object or potential target for range measurement, thereby reducing the pointing accuracy requirements.

[0024] As mentioned above, the vision system preferably controls the pointing of the LIDAR used to ping the one or more classified objects. The method may further include performing an additional pre-defined LIDAR scan pattern in parallel with the visually directed LIDAR pinging. This additional scan is preferably performed by the same random access LIDAR used to ping the classified objects. The additional pre-defined forward LIDAR scan (28 in FIG. 2b) is preferably along the direction of platform motion and operates to provide backup obstacle avoidance.

[0025] A simplified block diagram of a system for determining the location and / or navigation path of a mobile platform as described herein is shown in Figures 2a and 2b. The system includes a vision system 30 on a mobile platform 32 configured to identify an area of ​​interest 34 within a common visual / LIDAR FOV 35. The system includes a processor 36 configured to receive images captured by the vision system 30 and classify objects 38 within the area of ​​interest 34. The system also requires a random access LIDAR 40, which may be a coherent LIDAR configured to ping one or more of the classified objects 38. The processor 36 provides a list of points with coordinates corrected for the movement of the platform to the LIDAR 40, which provides corresponding range measurements to the processor 36. The processor 36 is further configured to determine the location of the platform 32 in three dimensions using data from the vision system 30 and the LIDAR 40.

[0026] The system is preferably configured so that the localization is performed continuously while the platform 32 is moving and / or the field of view (FOV) of the vision system 30 is changing. As described above, the vision system 30 may be configured to capture images of the region of interest 34 with known intervals between image captures. The system is further configured to define at least three non-collinear static points within the region of interest 34 to be used as common reference points for each image capture.

[0027] The processor 36 preferably uses at least one smart vision algorithm, such as a machine learning algorithm, to classify the object 38. By using a trained machine learning algorithm for navigation, the overhead of data collection and processing is reduced. Data defining several types of objects may be accumulated before the system is deployed, and the processor software can then maintain and update a library of objects of interest that are detected in real time over time. An "object of interest" is a physical object or part of an object that is important as a potential obstacle, target, or navigation feature for location determination within the environment of the mobile platform. As an old object of interest moves out of the FOV, a new reference point is added and used. The processor software may include a trained machine learning "classifier" algorithm that identifies and classifies objects of interest. The software may be configured to select reference features for location determination that are part of an object (e.g., a window or entrance of a building, a utility pole, etc.) from classes that are most likely to be stationary and have sufficient contrast.

[0028] The stationary reference features are confirmed to be stationary, for example, by comparing the measured platform motion with camera pixel motion and corresponding range measurements performed with a random access LIDAR between two or more time instances. The corresponding angular and linear displacements may be measured by an IMU. The relative angular position and distance from multiple (>3) stationary reference features may be used to determine the system's location and attitude.

[0029] The vision system 30 is preferably configured to operate in the visible and / or thermal portions of the spectrum. The LIDAR 40 is preferably configured to generate a diffraction-limited beam with an instantaneous field of view (IFOV) of 10μRad to 10mRad, a field of view of >10 degrees, and a repointing time of 1μs to 1ms.

[0030] The system allows for localization without an IMU. However, the system may further include an IMU 42 configured to determine the linear acceleration and / or angular acceleration and / or linear velocity and / or angular velocity of the platform 32. When so configured, the vision system 30, the LIDAR 40, and the IMU 42 are preferably time-synchronized using an external reference, such as an on-board clock or a GPS. The platform can generally be located periodically. The IMU 42 can be used to correct platform disturbance-induced errors in LIDAR pointing due to the time lag between camera image capture and LIDAR range data collection. This time lag is preferably <100 ms. Because IMUs tend to drift, the time between localizations, i.e. the time difference used to distinguish stationary from moving objects, is preferably small (<1 sec). These techniques can be used to provide feed-forward correction and line-of-sight stabilization, thereby enabling high-precision pointing in the presence of severe platform disturbances. Stabilized pointing allows for ranging of small reference features, thereby providing high localization accuracy in the lateral direction (orthogonal to the longitudinal line of sight).

[0031] A Global Positioning System (GPS) unit 46 may also be included, configured to update the absolute world position of the platform 32. The system may also include a Kalman filter 44 for estimating location and orientation using all available sensor measurements over a period of time, which may also receive input from one or more other sensors 45, such as a compass magnetometer.

[0032] The camera IFOV, which is the angular projection of a single camera pixel in the image, and the LIDAR resolution are preferably designed to have comparable vertical and horizontal distance measurement accuracy at the range limit of the LIDAR (e.g., nominal 2 cm at a distance of 100 m).

[0033] The present system and method for determining the location and / or navigation path of a mobile platform is particularly advantageous for use on small platforms (e.g., fixed-wing unmanned aerial vehicles) with challenging obstacle environments and platforms with size, weight, power, and / or processing power limitations. It can be more accurate than magnetometers and is not sensitive to electromagnetic fields. Additionally, the present invention is more compact and flexible than other physical sensors such as odometers. The present system does not require contact with a surface as odometers require, and does not require setting up an external reference source as in RF or ultrasonic triangulation based systems. Using selected features and random access LIDAR, the data and computational burden of the process is greatly reduced and the speed is significantly improved. The visual / LIDAR based localization described herein can be extremely accurate, with accuracy determined primarily by the accuracy of the LIDAR and the camera IFOV.

[0034] The embodiments of the invention described herein are illustrative and numerous modifications, variations, and rearrangements can readily be envisioned to achieve substantially equivalent results, all of which are within the spirit and scope of the invention as defined in the appended claims. The following items are elements of the claims as originally filed: (Item 1) 1. A method for determining a location and / or a navigation path of a mobile platform, comprising: using a vision system on the moving platform to identify the region of interest; classifying objects within the region of interest; instructing a random access LIDAR to ping one or more of the classified objects; identifying the location of the platform in three dimensions using data from the vision system and the LIDAR; A method comprising: (Item 2) each of the vision system and the random access LIDAR having a field of view (FOV); 2. The method of claim 1, wherein the vision system and the random access LIDAR are calibrated to establish a predetermined relationship between the respective FOVs. (Item 3) 2. The method of claim 1, wherein the pings performed by the LIDAR are used to determine the orientation of classified objects within the region of interest. (Item 4) 2. The method of claim 1, wherein the classifying, indicating and identifying steps are performed continuously while the platform is moving and / or while a field of view (FOV) of the vision system is changing. (Item 5) the vision system captures images of the region of interest with a known interval between image captures; The method further comprises: 5. The method of claim 4, further comprising a definition step of defining at least three non-collinear static points within the region of interest to be used as common reference points for each image capture. (Item 6) 2. The method of claim 1, wherein the classification step comprises the use of at least one smart vision algorithm. (Item 7) The method further comprises: defining at least three non-collinear static points within the region of interest to be used as common reference points for each image capture; 7. The method according to claim 6, wherein the classification step is used to select the static reference points. (Item 8) 7. The method of claim 6, wherein at least one of the smart vision algorithms is a machine learning algorithm. (Item 9) the LIDAR is a coherent LIDAR; The method further comprises: 2. The method of claim 1, further comprising a determining step of determining a relative velocity of the object with respect to the moving platform using the coherent LIDAR. (Item 10) The method further comprises: defining at least three non-collinear static points within the region of interest to be used as common reference points for each image capture; 10. The method of claim 9, wherein the relative velocities of the objects are used to select the static reference point. (Item 11) 2. The method of claim 1, wherein the vision system operates in the visible and / or thermal part of the spectrum. (Item 12) 2. The method according to claim 1, wherein the LIDAR has an instantaneous field of view (IFOV) of 10 μRad to 10 mRad. (Item 13) 2. The method of claim 1, wherein the LIDAR has a field of view of >10 degrees. (Item 14) 2. The method according to claim 1, wherein the LIDAR has a repointing time of "1 μs to 1 ms". (Item 15) Item 1. The method of item 1, wherein the LIDAR produces a diffraction-limited beam. (Item 16) The method further comprises: 2. The method according to claim 1, comprising a determining step of determining a linear acceleration and / or an angular acceleration and / or a linear velocity and / or an angular velocity of the platform. (Item 17) the vision system controls the pointing of the LIDAR used to ping one or more of the classified objects; The method further comprises: Item 17. The method of item 16, comprising using the linear acceleration and / or the angular acceleration and / or the linear velocity and / or the angular velocity to apply corrections to the LIDAR to maintain a stable pointing indicated by the vision system. (Item 18) the vision system controls the pointing of the LIDAR used to ping one or more of the classified objects; The method further comprises: 2. The method of claim 1, further comprising the step of performing an additional predefined LIDAR scan pattern in parallel with the visually indicated LIDAR pinging. (Item 19) 20. The method of claim 18, wherein the predefined LIDAR scan is along a direction of motion of the platform and acts to provide backup obstacle avoidance. (Item 20) 20. The method of claim 18, wherein the predefined LIDAR scan is used to effectively increase the LIDAR spot size of a detected object. (Item 21) 1. A system for determining a location and / or a navigation path of a mobile platform, comprising: a vision system on the moving platform configured to identify an area of ​​interest; a processor configured to receive images captured by the vision system and classify objects within the region of interest; a random access LIDAR configured to ping one or more of the classified objects; Equipped with the processor is further configured to determine a location of the platform in three dimensions using data from the vision system and the LIDAR. system. (Item 22) 22. The system of claim 21, further configured such that the determining of location is performed continuously while the platform is moving and / or while a field of view (FOV) of the vision system is changing. (Item 23) the vision system captures images of the region of interest with a known interval between image captures; The system further comprises: 23. The system of claim 22, configured to define at least three non-collinear static points within the region of interest to be used as common reference points for each image capture. (Item 24) 22. The system of claim 21, wherein the processor uses at least one smart vision algorithm to classify the object. (Item 25) 25. The system of claim 24, wherein at least one of the smart vision algorithms is a machine learning algorithm. (Item 26) 22. The system of claim 21, wherein the LIDAR is a coherent LIDAR. (Item 27) 22. The system of claim 21, wherein the vision system is configured to operate in the visible and / or thermal part of the spectrum. (Item 28) 22. The system of claim 21, wherein the LIDAR has an instantaneous field of view (IFOV) of 10 μRad to 10 mRad. (Item 29) 22. The system of claim 21, wherein the LIDAR has a field of view of >10 degrees. (Item 30) 22. The system of claim 21, wherein the LIDAR has a repointing time of "1 μs to 1 ms." (Item 31) 22. The system of claim 21, wherein the LIDAR is configured to generate a diffraction-limited beam. (Item 32) The system further comprises: Item 22. The system of item 21, comprising an inertial measurement unit (IMU) configured to determine linear acceleration and / or angular acceleration and / or linear velocity and / or angular velocity of the platform. (Item 33) 33. The system of claim 32, wherein the vision system, the LIDAR, and the IMU are time synchronized. (Item 34) The system further comprises: Item 34. The system of item 33 comprising a Kalman filter. (Item 35) The system further comprises: 22. The system of claim 21, comprising a Global Positioning System (GPS) unit configured to provide an absolute world position of the platform.

Claims

1. 1. A method for determining a location and / or a navigation path of a mobile platform, comprising: using a vision system on the moving platform to identify the region of interest; classifying objects within the region of interest; directing a random access LIDAR to ping one or more of the classified objects; identifying the location of the platform in three dimensions using data from the vision system and the LIDAR; A method comprising:

2. each of the vision system and the random access LIDAR having a field of view (FOV); 2. The method of claim 1, wherein the vision system and the random access LIDAR are calibrated to establish a predetermined relationship between the respective FOVs.

3. The method of claim 1 , wherein the pings performed by the LIDAR are used to determine the orientation of classified objects within the region of interest.

4. 2. The method of claim 1, wherein the steps of classifying, directing, and identifying are performed continuously while the platform is moving and / or while a field of view (FOV) of the vision system is changing.

5. the vision system captures images of the region of interest with a known interval between image captures; The method further comprises: The method of claim 4 , comprising a defining step of defining at least three non-collinear static points within the region of interest to be used as common reference points for each image capture.

6. The method of claim 1 , wherein the classifying step comprises the use of at least one smart vision algorithm.

7. The method further comprises: A definition step of defining at least three non-collinear static points within the region of interest to be used as common reference points for each image capture; The method of claim 6 , wherein the classification step is used to select the static reference points.

8. The method of claim 6 , wherein at least one of the smart vision algorithms is a machine learning algorithm.

9. the LIDAR is a coherent LIDAR; The method further comprises: The method of claim 1 , comprising determining a relative velocity of the object with respect to the moving platform using the coherent LIDAR.

10. The method further comprises: A definition step of defining at least three non-collinear static points within the region of interest to be used as common reference points for each image capture; The method of claim 9 , wherein the relative velocities of the objects are used to select the static reference point.

11. The method of claim 1 , wherein the vision system operates in the visible and / or thermal portion of the spectrum.

12. The method of claim 1, wherein the LIDAR has an instantaneous field of view (IFOV) of "10 μRad to 10 mRad."

13. The method of claim 1 , wherein the LIDAR has a field of view of >10 degrees.

14. The method of claim 1 , wherein the LIDAR has a repointing time of "1 μs to 1 ms."

15. The method of claim 1 , wherein the LIDAR produces a diffraction-limited beam.

16. The method further comprises: The method of claim 1 , comprising a determining step of determining the linear and / or angular acceleration and / or linear and / or angular velocity of the platform.

17. the vision system controls the pointing of the LIDAR, which is used to ping one or more of the classified objects; The method further comprises:

17. The method of claim 16, comprising using the linear acceleration and / or the angular acceleration and / or the linear velocity and / or the angular velocity to apply corrections to the LIDAR to maintain a stable pointing directed by the vision system.

18. the vision system controls the pointing of the LIDAR, which is used to ping one or more of the classified objects; The method further comprises:

2. The method of claim 1, comprising performing an additional predefined LIDAR scan pattern in parallel with the visually oriented LIDAR pinging.

19. 20. The method of claim 18, wherein the predefined LIDAR scan is along a direction of motion of the platform and acts to provide backup obstacle avoidance.

20. 20. The method of claim 18, wherein the predefined LIDAR scan is used to effectively increase the LIDAR spot size of a detected object.

21. 1. A system for determining a location and / or a navigation path of a mobile platform, comprising: a vision system on the moving platform configured to identify an area of ​​interest; a processor configured to receive images captured by the vision system and classify objects within the region of interest; a random access LIDAR configured to ping one or more of the classified objects; Equipped with the processor is further configured to determine a location of the platform in three dimensions using data from the vision system and the LIDAR. system.

22. 22. The system of claim 21, wherein the system is further configured such that the determining of location is performed continuously while the platform is moving and / or while a field of view (FOV) of the vision system is changing.

23. the vision system captures images of the region of interest with a known interval between image captures; The system further comprises:

23. The system of claim 22, configured to define at least three non-collinear static points within the region of interest to be used as common reference points for each image capture.

24. 22. The system of claim 21, wherein the processor uses at least one smart vision algorithm to classify the object.

25. 25. The system of claim 24, wherein at least one of the smart vision algorithms is a machine learning algorithm.

26. 22. The system of claim 21 , wherein the LIDAR is a coherent LIDAR.

27. 22. The system of claim 21, wherein the vision system is configured to operate in the visible and / or thermal portion of the spectrum.

28. 22. The system of claim 21, wherein the LIDAR has an instantaneous field of view (IFOV) of "10 μRad to 10 mRad."

29. 22. The system of claim 21, wherein the LIDAR has a field of view of >10 degrees.

30. The system of claim 21 , wherein the LIDAR has a repointing time of “1 μs to 1 ms.”

31. 22. The system of claim 21, wherein the LIDAR is configured to produce a diffraction limited beam.

32. The system further comprises:

22. The system of claim 21, comprising an inertial measurement unit (IMU) configured to determine linear and / or angular acceleration and / or linear and / or angular velocity of the platform.

33. 33. The system of claim 32, wherein the vision system, the LIDAR, and the IMU are time synchronized.

34. The system further comprises:

34. The system of claim 33 comprising a Kalman filter.

35. The system further comprises:

22. The system of claim 21, comprising a Global Positioning System (GPS) unit configured to provide an absolute world position of the platform.

Images