Tadej Jagodnik (2012) Detection of passengers in a car. EngD thesis.
Abstract
This thesis is focused on the research of the performance, suitability and reliability of some developed algorithms for face detection in the in-car environment. The vehicle interior is exposed to various illumination conditions which can complicate detection. Cramped space can also represent a problem, since it can cause occlusion. First we present three different approaches for face detection. To improve algorithm's performance and reduce the number of false detections, we suggest splitting frames or search window to two lines and five rectangles. We test algorithms on videos, recorded with wide-angle digital camera and annotate them by bounding boxes with implemented tool. System for evaluation is implemented to compare the algorithm's results with the results of ground truth annotation. Based on the evaluation system we perform various tests and with changing algorithm's parameters we try to determine their best performance.
Actions (login required)