Detecting Changes in Images of Street Scenes

Abstract

Abstract. In this paper we propose an novel algorithm for detecting changes in street scenes when the vehicle revisits sections of the street at different times. The proposed algorithm detects structural geometric changes, changes due to dynamically moving objects and as well as changes in the street appearance (e.g. posters put up) between two traversal times. We exploit geometric, appearance and semantic information to determine which areas have changed and formulate the problem as an optimal image labeling problem in the Markov Random Field framework. The approach is evaluated on street sequences from 3 different locations which were visited multiple times by the vehicle. The proposed method is applicable to monitoring and updating models and images of urban environments.

Street environment change detection from mobile laser scanning point clouds

“…Yin and Collins (2007) detect moving objects by a Belief Propagation approach using a 3D Markov Random Field (MRF). A similar method has been presented by Košecka (2013) to detect changes from street scene images. Changes are differentiated as structural, appearance change or temporary dynamically moving objects.…”

“…Laser scanning provides precise 3D geometric information on the environment, which is of great interest for 3D mapping, localization, scene perception, motion tracking and navigation purposes. Studies from computer vision mainly use imagery for city and street scene change detection (Pollard and Mundy, 2007;Sakurada et al, 2013;Košecka, 2013;Eden and Cooper, 2008;Taneja et al, 2011Taneja et al, , 2013. However, lidar (light detection and ranging) data (also referred to as laser scanning data, range data or lidar point clouds) have been proven to be an accurate data source for 3D urban reconstruction (Lafarge and Mallet, 2011;Chauve et al, 2010;Verma et al, 2006;Zhou and Neumann, 2010;Toshev et al, 2010;Banno et al, 2008;Poullis, 2013), infrastructure management and road inventory (Pu et al, 2011; Vosselman, 2012).…”

Geometric Change Detection in Urban Environments Using Images

“…[16] instead proposed to detect changes based on the appearance and disappearance of 3D lines detected in the images. [17] also compares two set of images captured at different time instants by first recovering a coarse structure of the scene. The images are segmented into superpixels for this purpose and the corresponding superpixels across the old and new images are compared to reveal changes.…”

Fault-Diagnosing SLAM for Varying Scale Change Detection

“…For PC, we introduce a SIFT feature-based PC approach. Based on the literature (e.g., [1]), the LoC of a query live feature is measured according to its dissimilarity to the most similar normal feature. Firstly, every live/map image is represented as a collection of SIFT features with Harris-Laplace keypoints [13].…”

“…For long-term map maintenance in dynamic environments, a robotic visual SLAM system must detect changed objects (e.g., furniture movement, and building construction) in a live image with respect to the map, while ignoring nuisance changes (e.g., sensor noises, registration errors, and occlusions) during long-term multi-session navigation. One approach is to formulate the problem as a pair-wise image comparison (PC), to compare each live-map-image-pair using image differencing techniques [1]. However, this requires that a robot memorizes every map image; hence, scaling to large-size environments is difficult.…”