[Pycon] [new paper] "Andrea Benericetti" - Machine learning approaches for road scene video analysis

[info a pycon.it]

Title: Machine learning approaches for road scene video analysis
Duration: 45 (includes Q&A)
Q&A Session: 15
Language: en
Type: Talk

Abstract: Dashboard cameras (dashcams), inward or front-facing cameras installed in personal or commercial vehicles, are becoming increasingly popular due to the pervasiveness of their applications: driver safety, autonomous driving, fleet management systems, insurance, theft detection are some examples. Focusing on safety, one of the main problems is to analyze videos and automatically detect dangerous situations occurring in them, such as the risk of a near crash with another vehicle or pedestrian.
In this talk, we show how we tackle this real-world problem at Verizon Connect, using a mix of state-of-the-art deep learning methods and traditional computer vision / machine learning techniques, leveraging libraries such as scikit-learn, scipy/numpy, pandas, openCV, and keras/tensorflow. We also describe how we use AWS and docker to deploy, serve and scale the application to customers all over the world.

Tags: [u'ComputerVision', u'scikit-learn', u'opencv', u'optical-flow', u'video', u'machine-learning', u'analytics', u'pandas']