Názov:Predicting Human Behavior from Public Cameras with Convolutional Neural Networks
Vedúci:prof. Ing. Igor Farkaą, Dr.
Kµúčové slová:pedestrian movement predictions, human behavior prediction, crowd analysis, deep learning, neural networks, convolutional neural networks, convolutional encoders
Abstrakt:Prediction of human behaviour and crowd analysis is one of the most important applications of computer vision. Recently, deep learning methods have been shown to produce interesting results when applied to computer vision problems. In our work we investigate possible application of these methods to the problem of predicting a pedestrians movement from the video captured by surveillance camera capturing public space, such as metro, train or bus stations. We used convolutional neural networks to predict a future mask of pedestrians locations. In our approach we focus on prediction of these locations from raw video frames where our models do not have any prior information about each pedestrian movement. We present our results and provide extensive error analysis of trained networks and the problem as such.

Súbory diplomovej práce:

Predicting Human Behavior from Public Cameras with Convolutional Neural Networks.pdf