Data Dimensionality (Number of Objects) Conventional Sensor Border Preprocessing Sensor Border Regular Flow High Computational Demand Less Regular Flow Lower Demand * Image Capture * Spatial Filtering * Edge/Motion Detection * Decision Making * Algorithm Control * Conditional Jumps * Image Segmentation * Object Labeling * Feature Extraction Irregular Flow Moderate Demand Abstraction Level (Data Structure Complexity) (a) ~28 K bytes Amount of Data ~4 K bytes ~4 K bytes <100 bytes Processing Results <10 bytes Area, Perimeter Number of Holes, Major Axis, Extent... Yes/No Blob Analysis and Features Extraction (ROI) Classifier Local Threshold HDR Acquisition Processing Chain Morphological Processing Th High Bandwidth LP-Filter (b) Figure 3. (a) Processing hierarchy, from left-top to bottom-right, in vision; (b) Illustrative example of vision processing chain [19]. This figure highlights the steps to go from sensor raw data to vision outcomes. The vertical axis represents data dimensions while the horizontal one represents the abstraction level of the data. The processing chain follows 94 IEEE cIrcuIts and systEms magazInE the diagonal arrow in Fig. 3(a). Top-left corresponds to input data captured by the sensor and bottom-right corresponds to output data which support system actions. The first stage of the vision processing chain is usually sEcOnd quartEr 2018