Objectives: Detecting arbitrary moving objects in the scene
Needs: Improving safety in case that deep learning-based object detection fails or moving objects are not in traing datasets
Method: 3D Computer Vision (Epipolar Geometry)
Issues: Hard to tell if the objects seems to move because the vehicle is moving or they are moving by themselves
Objectives: Combining different sensors to make SLAM more robust to challenging environments such as wide-open areas and corridors
Sensors
LiDAR: Geometric information
Camera: Texture information
IMU Sensor: Ego-centric motion
GPS Sensor: Absolute position
Method: Continuous-time trajectory to combine asynchronous sensor data
Objectives: Tracking the camera pose in real-time using an RGB-D camera to overlay virtual objects on augmented reality glasses
Method: Using the depth uncertainty as weights of the depth residuals in optimization based on the sensor model of the RGB-D camera
Results: Real-time camera pose tracking more robust to intensity change and depth noise