Fusion-Based Background-Subtraction using Contour Saliency

Abstract

We

Structure from Infrared Stereo Images

“…For each stereo pair, a ground-truth disparity map is also constructed for the foreground objects of the scenes to be used in our experiments. The reason that we only consider the disparity of foreground objects in our quantitative evaluation is that background objects in indoor environments are usually at "thermal crossover" (i.e., thermal properties of the objects are relatively similar to those of the surrounding environment [17]), and therefore the captured IR images at those areas do not contain enough information for image processing tools. …”

“…Davis and Sharma [17] proposed a fusion technique assuming that the two thermal and IR images are then registered and fused together, to produce the final results. The system has proven to be able to detect more than 95% of pedestrians up to 45m and more than 89% up to 75m.…”

“…Therefore, systems based on the assumption that pedestrians are hotter than the environment usually produce partly or complete misdetections. In order to cope with these deficiencies, some pedestrian detection [17], [6] and tracking [52] techniques have been proposed which are based on contemporary use of a visible and an infrared system in order to utilize the benefits of both approaches. Although, only a few research efforts have been carried out on this subject, the result of visible and infrared fusion has been relatively promising.…”

Registration of Multimodal Stereo Images Using Disparity Voting from Correspondence Windows

“…Specifically the disparate, yet complementary nature of visual and thermal imagery has been used in recent works to obtain additional information and robustness [1], [7]. The use of both types of imagery yields information about the scene that is rich in color, depth, motion and thermal detail.…”

“…In long range surveillance applications [7], the cameras are assumed to be oriented in such a way that a global alignment function will register all objects in the scene. However, this assumption means that the camera must be very far away from the imaged scene.…”