Risk assessment from video data KEY ASPECTS Risk

Risk assessment from video data KEY ASPECTS Risk assessment on roads (focus on developing countries) Exploit video analysis techinques Integrate open data resources LIP – University of Lisbon – November 2 nd 2018 1

Risk assessment from video data § Goals: - Define a protocol to assess risk factors in roads from video images and open data. - Use commodity hardware as much as possible. - Maximize automation of the procedure. 3. Assess risk level on portions of the chosen road (green, yellow, orange, red & black) 2. Build a score by combining the different values of indicators 1. Identify risk indicators that can be evaluated with (almost) no human intervention LIP – University of Lisbon – November 2 nd 2018 2

Risk assessment from video data Camera videos + GPS Road. Lab. Pro app (for road pavement condition) Open. Street. Map maps + metadata (for crossroads and road area) LIP – University of Lisbon – November 2 nd 2018 3

Risk assessment from video data Deep learning + image analysis techniques allow to detect the presence of persons and road signs… …but also road lines to some extent (and from those the curvature of the road) LIP – University of Lisbon – November 2 nd 2018 4

Risk assessment from video data The result has been a tool to make the data processing mostly automated: videos are uploaded, together with some manual metadata (name of the road, some camera calibration params, etc. ) and the risk analysis is performed. LIP – University of Lisbon – November 2 nd 2018 5

Video data and health Similar techniques have been used in healthcare context Component 1: find in a movie the presence and the position of human faces (Open. CV library) + Component 3: detect anomalies in the heartbeat frequency and recognize disease “footprints” LIP – University of Lisbon – November 2 nd 2018 Component 2: extract heartbeats from variations of the red channel inside the boxed areas of the image. + 6

CRM and marketing KEY ASPECTS Enrich customer DBs Social “behavior” profiling Choosing campaigns and contact channel per-customer LIP – University of Lisbon – November 2 nd 2018 7

CRM and marketing § Goals: - Acknowledge that digital connections and social networks have de facto changed completely face to marketing actions towards customers. - Replace traditional campaigns with customized advertising based on customer habits. - Exploit all available info (in the limit of privacy laws!) about single customers and similar profiles to find what products can be of interest. LIP – University of Lisbon – November 2 nd 2018 8

CRM and marketing name address job health real estate market salary … customer info context info … activities on brand pages activities on competitor pages goal info via customer segments (age, geo, etc. ) OPEN DATA LIP – University of Lisbon – November 2 nd 2018 9

CRM and marketing CUSTOMER INFO PAST PRODUCTS TOUCH POINT NEXT BEST PRODUCT LIFE CYCLE Bayesian models can synthetize all different data sources in a consistent probability evaluation of NBP. SOCIAL ACTIVITIES An user interface through web dashboards allows CRM decision makers to choose the most convenient marketing activity in each area. For privacy reason all displayed data are simulated. LIP – University of Lisbon – November 2 nd 2018 10

Further applications Similar techniques have been used in other problems related to profiling % % frodi, churn, ecc. Prob manutenzione predittiva LIP – University of Lisbon – November 2 nd 2018 11

Thank you for your attention LIP – University of Lisbon – November 2 nd 2018