Template-guided inspection of arbitrarily oriented targets

The performance of template-based inspection of randomly oriented targets relies on the effective operation of two procedures: image registration and change detection. Each procedure is characterised by its own challenges. Real-time applications require that image registration is performed in an acceptable time with accurate results, regardless of structural distortions. Change detection can be hindered by noise fluctuations and illumination variations. An additional challenge in template-guided change detection comes from the fact that the decision needs to be based on a comparison restricted to two frames. The operation principles of a template-guided change detector are analysed. Image registration is based on a wavelet-based method which employs mutual information. A block-based change detection algorithm, which separates content alterations from noise-level fluctuations, illumination variations and shadows, is also analysed. The principles presented are combined for the implementation of a sample template-guided undervehicle surveillance system. The architecture of the system and its experimental results are presented, and practical aspects and considerations regarding the system design are discussed.