Tests on Two Popular VIO Algorithms
Test date: July 31st.
Algorithms: MSCKF-VIO & VINS-MONO
Environment: UAV Lab (E4A 03-04, NUS)
Sensors: 2 Point Grey Cameras (Chameleon3 2.8mm) / VN-100 IMU (adis16480)
We test on two popular open source vio (visual-inertial odometry) algorithms, VINS-MONO and MSCKF-VIO. Apparently the major diference between them is the numbers of cameras, only one camera used in  but two in . Accroding to the ’s authors that stereo cameras could raise the performances, we did a simple test and therefore compared the results.
Install the camera and IMU drivers.
Implement two algorithms through their official open source ros package.
Calibrate the sensors.
Record a dataset in rosbag type.
Test two algorithms on the dataset.
Compare the results.
Notices When Test
Since  has the loop-closing mode that we need to turn it off.
There is a time-delay between imu and images which is short but need to be concerned about. The authors of  provide the delay calculation and we consider this delay is approximately accurate to use in .
 needs a step of raising vertically and the dataset must contains some rotation or accelerating to fulfill the experiment procedure.
The test video are showing below.
We record the odometry of these two algorithms and plot them in MATLAB in the same coordinate. Fig 1 shows that after turnning off the loop-closing mode,  has a drift in X-Y plane though we started and ended in the same position while the Fig 2 shows that  has a bigger drift in Z.
Besides,  needs more CPU resources because of the processing of stereo images. And  performs not so good when pirouetting.
 Qin, Tong, Peiliang Li, and Shaojie Shen. “VINS-Mono: A robust and versatile monocular visual-inertial state estimator.” IEEE Transactions on Robotics 99 (2018): 1-17.
 Sun, Ke, et al. “Robust stereo visual inertial odometry for fast autonomous flight.” IEEE Robotics and Automation Letters 3.2 (2018): 965-972.