| Summary: | 碩士 === 國立交通大學 === 多媒體工程研究所 === 102 === People localization is prevailing in applications of video surveillance and digital life. Due to the ever-increasing demand, it is necessary to improve its efficiency. While vanishing point-based line sampling is used to develop real-time multi-camera people localization methods, such methods need to obtain vanishing point of vertical lines for line sampling. Conventionally, the vanishing point is estimated by intersection of calibration pillars that are perpendicular to the ground plane. However, it may not be suitable to set calibration pillars in crowded scenes like street crossing. Accordingly, a pedestrian image-based algorithm is proposed in this paper which focuses on finding virtual calibration pillars from foreground (human) regions to avoid the difficulty of setting calibration pillars, making vanishing point-based line sampling methods easy-to-use.
|
|---|
