Inertial Odometry on Handheld Smartphones

Transcription

1 Inertial Odometry on Handheld Smartphones Arno Solin 1 Santiago Cortés 1 Esa Rahtu 2 Juho Kannala 1 1 Aalto University 2 Tampere University of Technology 21st International Conference on Information Fusion July 12, 2018

2 Introduction Phones have accelerometers and gyroscopes Should in theory enable inertial navigation Cheap and small sensors Low quality data inertial navigation not possible We demonstrate that it can be done 2/20

3 Inertial navigation: How it should work Inertial navigation Velocity is the integral of acceleration. Velocity is the integral of acceleration. Position is the integral of velocity. Position Weis can the observe integral of acceleration velocity. and angular velocity in the mobile phone. We can observe acceleration and angular velocity in the mobile phone. Position Velocity Acceleration 3/20

4 Inertial navigation: Why it does not work All inertial navigation systems suffer from integration drift. Small errors in the measurement of acceleration and angular velocity... Progressively larger errors in velocity... Even greater errors in position. The dominating component in acceleration is gravity. Even slight error in orientation makes the gravity leak. The sequential nature of the problem makes the errors accumulate. 4/20

5 Inertial navigation: How to make it work Input: accelerometer data a k and gyroscope data ω k. Accelerometer and gyroscope biases part of the state: ã k = T a k a k b a k ω k = ω k b ω k Dynamical model: p k p k 1 + v k 1 t k v k = v k 1 + [q k (ã k + ε a k )q k g] t k q k Ω[( ω k + ε ω k ) t k]q k 1 for position p k, velocity v k, and orientations q k at time t k. Inference by an Extended Kalman filter / smoother 5/20

6 Inertial navigation: How to make it work Additional constraints (observations) are required This framework can use Zero-velocity updates (ZUPTs) Position fixes Loop-closures Barometric air pressure for relative height A pseudo-measurement keeping the velocity component from exploding Sensor timing info A matter of learning the biases 6/20

7 How does this compare to previous approaches? On mobile devices PDR typically based on: Movement detection Step and heading systems Visual features All of these are limited in some way 7/20

8 Example studies Equipment used: Off-the-shelf iphone 6 Sensors: Built-in gyroscope and accelerometer Sampling rate: 100 Hz Computations: Off-line (runnable on device hardware) 8/20

9 Example: Conventional PDR 9/20

10 Example: Conventional PDR z-displacement (m) Bag Position fix Pocket Position fix Velocity (m/s) x y z Orientation (rad) Time (seconds) Figure 2: The altitude (vertical) profile, the velocities, and the orientations of the phone along the path in Figure 1. The shading shows the stationarity detection outcome, where zero-velocity updates were triggered. The subtle periodicity in the path is due to walking, and the drop in altitude on the first floor is because of the phone being in a bag for parts of the path. tic filtering theory can be applied to estimation of the double-integrated accelerometer readings for inferring 10/20 change of position over time. The extensive survey of Harle [17] covers many approaches with di erent con-

11 Example: Conventional PDR Vertical (z) displacement (meters) Floor level 3 Floor level 2 Floor level 1 Stairs Elevator ride Horizontal (y) displacement (meters) Figure 3: A PDR example with first descending two levels down and then taking the elevator back up. The path was started at origin and the path ends with a loop-closure in the same place. The points where the phone touches the floor level (the sharp drops in vertical position) are zero-velocity updates. No absolute position info was given to the model. 4. EXPERIMENTS This section is dedicated to versatility, and performance of system presented in the previou In the examples, the interest w and tablets mostly because of and good software compatibility device models used in the exam and the ipad Pro (12.9-inch mod are equipped with built-in IMU and a barometric sensor (Bosch all experiments, the IMU sensor timestamps were collected at 1 eter data (when used) at appro data was collected using an in-ho lection application, and the pa o -line in Mathworks Matlab. 4.1 Pedestrian Dead-Re The most apparent use case f is to apply it to pedestrian dea mobile phone (iphone 6) is carr There exist a multitude of meth using data provided by mobile aim of this experiment is to sho fers from others by its generalit 11/20 Figures 1 and 3 summarize fe INS system; the example incl

12 t =18.7 t =37.0 Example: General dead-reckoning iphone 6 Baby Freely turning front wheels Figure 4: The experiment setup with a wheeled baby stroller and an iphone placed on the stroller for tracking. altitude on the first floor is due to the phone being in the bag for a part of the path. The figure also shows the estimated velocity. The periodicity is due to walking. This e ect is less evident when the phone is in the bag, /20 t =55.8

13 Example: General dead-reckoning The video is available on YouTube: 13/20

14 Example: Inertial measurements The video is available on YouTube: 14/20

15 Example: Inertial measurements 7.36 m 7.36 m 8.35 m 8.41 m 8.33 m 8.47 m 7.42 m 7.39 m (a) Measurement #1 (b) Measurement #2 Figure 6: Two examples of measuring the wall placements of an indoor space with an ipad Pro. The walls (the equations of the planes) are a part of the state variable. Points where the phone is stationary (against the wall) are visualized in red. The true size of the room is 7.30 m 8.45 m. manoeuvred wheeled object with an intrinsic noise source that is a baby stroller with a baby on board. Figure 4 shows the test setup, where the phone is placed on the stroller leaning on the top. The phone (iphone 6) remains fixed to the stroller body for the entire experiroom. Associating several ZUPTs to the same wall with the additional knowledge that the points span a plane through the space, can also make Solin, it possible Cortés, Rahtu, to Kannala better estimate the wall placement and orientation. 15/20 The model in this paper is flexible enough that

16 Recap Inertial navigation on a smartphone is possible The key is learning the sensor transformations (biases) Not possible without measurements (consraints) ZUPTs pseudo-speed Future work towards relaxing these further 16/20

17 Homepage: 17/20