Detekce anomálií v chování řídiče

Abstract

The main topic of this work is the detection of anomalies in driver’s behavior. It describes technical difficulties of the problem of anomalous actions detection in the video. It summarizes the main methodologies generally used in the anomaly detection area. The core part of the work is the design and the experimental realization of the system that uses subsystem for keypoint detection in the video and deep neural networks to distinguish normal and anomalous actions of the driver. Several experiments are performed to validate the usability of selected keypoint features and neural network architectures.