Object detection and tracking is a challenging problem in the dynamic construction environment, especially for objects with kinematic joints and changing poses, e.g. excavators. This paper presents a kinematic key nodes model for dynamically detecting and locating movable objects in construction sites using color-depth cameras. The key nodes model is designed by first analyzing kinematic constraints from mechanic specifications of interested construction site objects and then building representative segmented contour templates under supervision. Feature matching is realized by comparing the real object components and segmented contour templates in a traversal order of the tree structure for all the key nodes. Since color-depth cameras capture not only images, but also the distance between detected objects and the camera itself in a format of three-dimensional (3D) point clouds, the 3D spatial location of an object is obtained by linking the image coordinate of each detected key node with the 3D point coordinates. This proposed framework showed great potential in construction monitoring and safety management.
展开▼