Multi-objective Mapping and Path Planning using Visual SLAM and Object Detection

摘要

Path planning of the autonomous robots is one of the crucial tasks that need to be
\nachieved for mobile robots to navigate through the environment intelligently. The robot
\npaths are typically planned utilizing map that is accessible at the time with a certain optimization
\nobjective such as to minimizing the travel distance, or time. This thesis proposes
\na multi-objective path planning approach by integrating Simultaneous Localization And
\nMapping (SLAM) with a graph based optimization approach and an object detection algorithm.
\nThe proposed approach aims not only to nd a path that minimizes travel distance
\nbut also to minimize the number of obstacles in the path to be followed.
\nThis thesis uses Visual SLAM (VSLAM) as the basis to generate graphs for global path
\nplanning. VSLAM generates a trajectory network which is usually in the form of a spare
\ngraph (if odometry based) or probabilistic relations on landmark estimates relative to the
\nrobot. An object detection algorithm is run in parallel to provide additional information
\non trajectory network graphs generated by the VSLAM, to be used in multi-objective
\npath planning. The VSLAM, object detection, and path planning elds are typically
\nstudied independently, but this thesis links the these elds to solve the multi-objective
\npath planning problem.
\nThe rst part of the thesis presents the connections and methodology on using the VSLAM
\nand object detection to generate trajectory network graphs. The nodes are inserted
\nto the graph when a new keyframe is needed in VSLAM. The distance travelled between
\nthe nodes is the rst criterion to minimize and is computed while traversing. In parallel
\nto VSLAM, the object detection component quanti es the number of objects detected
\nbetween the nodes. Only the pre-trained objects to detect are quanti ed and the trained
\nobjects in the thesis are cars and trucks. The number of objects are the two additional
\nedge information added to the graph. Later in the thesis, the multi-objective path planning
\non the generated graphs is presented. The objective of path planning on graph is not
\njust on minimizing the distance to travel but also on minimizing the number of cars and
\ntrucks it passes. The proposed design is tested using KITTI dataset which is specialized
\nfor autonomous driving and consists of many cars and trucks. The design is not limited to
\nautonomous driving applications, but can be applied to other elds such as surveillance,
\nrescuing, and many more with di erent objects to detect.