Artificial Neural Networks for 3-D Motion Analysis-Part I: Rigid Motion

This paper proposes a new approach applying artificial neural network techniques to three-dimensional (3-D) rigid motion analysis based on sequential multiple time frames. The proposed approach consists of two phases: 1) matching between every two consecutive frames and 2) estimating motion parameters based on the correspondences established among all the frames. Phase one specifies the matching constraints to ensure a stable and coherent feature correspondence establishment between two sequential time frames and configures a two-dimensional (2-D) Hopfield neural network to enforce these constraints. Based on the feature correspondences established among all the frames, phase two constructs a three-layered neural network to estimate the motion parameters through a supervised learning process. The proposed approach differs from the previous works in several respects. First, it tackles the problem of motion analysis based on sequential multiple time frames. Conventional motion analysis, by contrast, is mostly based on two frames due to the complexity in computation when the number of frames increases. Second, this approach represents an effective way to achieve optimal matching solution between two frames using neural network techniques. The energy function of the Hopfield network is designed to reflect the matching constraints and the minimization of this function leads to the optimal feature correspondence establishment. Third, this approach introduces the learning concept to motion estimation. The structure of the proposed learning neural network provides the flexibility in estimating motion parameters based on information from multiple frames. Finally, this approach contributes to research fields of both motion analysis and neural network by presenting a new and fast-implemented framework for motion analysis based on multiple frames and proposing a unique neural network architecture and dynamics. Several experiments on both synthetic and real data are conducted to corroborate the proposed approach.