Perception Team- SLAM for home robot vacuum- Indoor Semantic Mapping- Depth data based indoor obstacle detection.

AR Lab of Baidu AI Group.- Researched and developed 3D model localization and tracking prototype with the moving edge method. Applied a particle filter model during tracking and a simulated annealing step during initialization. Integrated with other methods as 3D model tracking and pose estimation module for our AR SDK.- Using ARKit API, developed an iOS client tool collecting and sending ARKit tracking data with Protobuf for our Visual Positioning System experiments.

- Developed mobile Mixed Reality engine and visual odometry system. Responded for 2D target detecting and tracking module. With given image template, selected feature based 2d-2d correspondence during initialization and solved the optimized camera pose. During the tracking step, applied the Optical Flow method to track key points and updated poses with the nonlinear optimizer. The re-projection error is used as the metrics threshold for entering re-initialization.- Launched iOS AR game MuzzBloc. With CNN model, detected and classified objects in the already segmented image bounding boxes. Used features correspondence with image template for target localization and tracking.- Demo: https://www.muzzbloc.com/

- Researched adaptive sampling and reconstruction problems in Monte Carlo rendering in Computer Graphics.- Presented a novel method sampling ray adaptively utilizing spatial and material features of the scene. Developed these work into my bachelor thesis.