Auto-organized visual perception using distributed camera network

Abstract

Abstract. Camera networks are complex vision systems difficult to control if the number of sensors is getting higher. With classic approaches, each camera has to be calibrated and synchronized individually. These tasks are often troublesome because of spatial constraints, and mostly due to the amount of information that need to be processed. Cameras generally observe overlapping areas, leading to redundant information that are then acquired, transmitted, stored and then processed. We propose in this paper a method to segment, cluster and codify images acquired by cameras of a network. The images are decomposed sequentially into layers where redundant information are discarded. Without need of any calibration operation, each sensor contributes to build a global representation of the entire network environment. The information sent by the network is then represented by a reduced and compact amount of data using a codification process. This framework allows structures to be retrieved and also the topology of the network. It can also provide the localization and trajectories of mobile objects. Experiments will present practical results in the case of a network containing 20 cameras observing a common scene.

Efficient Feature Distribution for Object Matching in Visual-Sensor Networks

“…Patricio et al [27] applied a multiagent framework to a VSN to control the capture parameters of a surveillance system and Gilbert and Bowden [28] presented a system for the incremental, scalable, inter-camera tracking of objects. In the work of Chang et al [29] a camera network system can estimate its topology and auto-organize its own activities according to the content of the scene and the task to be undertaken. This is achieved by using custom encoding, which reduces the amount of information that has to be transmitted across the network.…”

Adapting a real-time monocular visual SLAM from conventional to omnidirectional cameras

“…Their experiments show that active windows provide best matching results than square windows. Ieng et al [5] propose the computation of patches of different angular apertures for the same feature to overcome the matching problems derived from the varying resolution of the camera. Scaramuzza et al [21] take advantage of the projection of vertical lines of the world as radial lines on the image.…”

Geometric and probabilistic image dissimilarity measures for common field of view detection

“…Typically, local feature correspondences are used to identify images with a common field of view [11,2]. Snavely et al [11] use SIFT features [7] with strict rejection in case of ambiguities during the matching process.…”