Change detection in streetscapes from GPS coordinated omni-directional image sequences

Abstract

As part of ITS technology, to achieve quick map updates, we propose a method for automatically detecting changes in streetscapes from images captured by car-mounted omnidirectional cameras. It comprises two stages; accurate alignment of a map and street images taken at various times, and detection of changes in streetscapes from the aligned data. The system will collect data via many free-running cars fitted with low-cost equipment to obtain images at various times and along routes. In the first stage, we process the alignment of the image frames taken at same locations and determine the accurate position information of each frame by a method composed of dimension reduction and DP matching. Then in the second stage, we detect changes in streetscapes from images taken at various times. Experiments with 44 data items which were collected over about a year, demonstrate the effectiveness of our method.

Single camera vehicle localization using SURF scale and dynamic time warping

“…They also usually employ supporting sensors to achieve localization-either an IMU [5] or odometry information [14]. A simpler approach is to localize against the closest database image, of known location, using DTW (or Dynamic Programming) to remove temporal differences between input and database image streams [6], [15]. The matching between sample images and those in the database can be performed by using a kind of template matching [15], or by using a low bit-rate image sequence instead of single images [16], which is stable in varying environments but does not provide high localization accuracy.…”

“…A simpler approach is to localize against the closest database image, of known location, using DTW (or Dynamic Programming) to remove temporal differences between input and database image streams [6], [15]. The matching between sample images and those in the database can be performed by using a kind of template matching [15], or by using a low bit-rate image sequence instead of single images [16], which is stable in varying environments but does not provide high localization accuracy. Novel feature based methods have also been proposed [7], where the epipole calculated from matched features between images is tracked.…”

Scene segmentation of wedding party videos by scenario-based matching with example videos

“…The DTW method is a sequence matching method that could be applied to two sequences with different lengths, and has also been used for video segment matching [2,4,6]. When two video segments X and Y are represented as sequences of feature vectors (x 1, .…”

Smart Phone Picture Recognition Algorithm Using Electronic Maps of Architecture Configuration

“…피사체 인식 방법은 마크를 사용하는 방법 [16,28], 이미지 데이터베이스에 있는 이미지 들과 비교하는 방법 [8,23], 모델 기반 방법 [22,26], 랜드마크(landmark) 기반의 방법 [20,25] …”