A morphological approach to moving-object recognition with applications to machine vision.

摘要

A methodology based on mathematical morphology is proposed for both shape recognition and motion estimation of two-dimensional (2-D) objects or shapes. This novel approach is based on the introduction of a shape descriptor called the Morphological Autocorrelation Transform or MAT. The MAT of an image is composed of a family of Geometrical Correlation Functions (GCFs) which define the morphological covariance in a specific direction. The MAT is shown to be translation-, scale-, and rotation-invariant. Also, in most situations, a small subset of the MAT suffices for image representation.; First, the characteristics and performance of a shape-recognition system based on the MAT are investigated and analyzed. A criterion based on the area under the GCF curve provides promising results. Computational complexity of the proposed system is examined. It is shown that important shape properties, such as area, perimeter, and orientation, are readily derived from the MAT representation.; Second, a new algorithm for motion-parameter estimation based on the family of GCFs is developed. Under relatively weak conditions, it provides a very fast and effective way of estimating the speed and direction of a moving-object. Its computational complexity is studied. Experimentally, it is shown to work well for relatively fast-moving objects.; Third, new high-speed architectures are proposed for efficient realization of the proposed schemes. Specifically, a Nonlinear Pipeline Processor (NPP) has been created to implement the two basic morphological transformations: dilation and erosion. NPP, a basic building block which can be used to realize the MAT, is highly modular and well-suited for VLSI implementation.; Finally, the proposed integrated scheme is applied to the conveyor-belt problem in flexible automation. Experiments performed in the Computer-Integrated-Manufacturing (CIM) Laboratory of the Department of Mechanical Engineering at the University of Toronto demonstrated that the proposed algorithm works very well in spite of noise, motion blur and mechanical vibration.