Object Tracking With Multiple Non-overlapping Cameras

With the development of society, intelligent monitoring technology has gradually become the focus in computer vision field. In order to meet the needs of monitoring environment and obtain the moving object’s comprehensive information, we need multi-cameras’collaboration. Due to the factors of monitoring environment’scope and the amount of calculation, the application of multi-cameras monitoring technology with non-overlapping filed of view (FOV) is more and more widely.In multiple non-overlapping cameras’monitoring environment, as to the influence of different lighting conditions and the object’s posture in different cameras’FOV, making the same object has different images in different cameras, and then making the object matching difficulty between multi-cameras. Also, because of the presence of blind areas among multi-cameras, when the object disappears in the blind area, the system can not get the object’s moving information and can’t achieve the continuous tracking. This paper has done the study about object tracking with multiple non-overlapping cameras. The main contents are as follows:First, to achieve the object detection and tracking under a single camera. We have done some improvements of shadow removal and object tracking algorithms.Second, aim to the question of object matching in multiple non-overlapping cameras, we proposes two methods of object matching:(1) In the case of having a training set, this paper proposes a layered object matching method using the brightness transfer function(BTF) in a low dimensional subspace. We use the probabilistic principal component analysis (PPCA) to reduce the BTF space’s dimension and take the bidirectional mapping, coupled with the object color features’distribution to do the object matching.(2) Without a training set. In a quantifized HSV color space, we use the information entropy to describe the contribution of an object’s color distribution to its characteristic, which adds to the distribution information in object’s color characteristics, and then improving the object matching’s accuracy.Third, our paper puts forward an object tracking method based on Agent with multiple non-overlapping cameras, where each camera is represented and managed by an intelligent and collaborative agent to solve complex environment problems, such as occlusion, blind areas existing in multi-cameras. What’s more, the camera-agents can collaborate with each other through agents’communication mechanism to improve global cooperation.Fourth, use JADE platform to build Multi-Agent Systema, and intial realized the Multi-Agent Cameras System (MACS). Using real video data to test our method which based on Agent object tracking is effective.