Egomotion estimation and the detection of moving objects with delayed-type CNN templates

Spatial-temporal event detections are crucial tasks in machine-vision and they are usually difficult to be handled efficiently with current algorithms and devices. In this article we show how cellular neural networks with delayed type templates are capable of detecting certain spatial-temporal features and how these features can be used for simple egomotion estimation. The detection and estimation is done by using continuous dynamics without cutting the input flow into frames. We can observe similar structures -the analogy of delayed type templates- in the retina, which performs well and efficiently in image processing tasks. Delayed type templates can provide us with even more flexibilities and possibilities in new applications including frameless detection of motion features.