An Adaptive Background Model for Camshift Tracking with a Moving Camera

Abstract

Continuously Adaptive Mean shift (CAMSHIFT) is a popular algorithm for visual tracking, providing speed and robustness with minimal training and computational cost. While it performs well with a fixed camera and static background scene, it can fail rapidly when the camera moves or the background changes since it relies on static models of both the background and the tracked object. Furthermore it is unable to track objects passing in front of backgrounds with which they share significant colours. We describe a new algorithm, the Adaptive Background CAMSHIFT (ABCshift), which addresses both of these problems by using a background model which can be continuously relearned for every frame with minimal additional computational expense. Further, we show how adaptive background relearning can occasionally lead to a particular mode of instability which we resolve by comparing background and tracked object distributions using a metric based on the Bhattacharyya coefficient.

An EM/E-MRF algorithm for adaptive model based tracking in extremely poor visibility

“…[46][47][48]. The ABCshift algorithm [49], also completely relearns the background model at every iteration, but uses a static object model and does not make use of prior knowledge of the 3D structure of the tracked object.…”

Construction of a Video Dataset for Face Tracking Benchmarking Using a Ground Truth Generation Tool

“…Furthermore, various types of head motions will affect the accuracy of these tracking algorithms. The second approach is to define the location of the facial area, based on the skin color of faces [7], [8], [19]. This can deal with fast movements, occlusion and scale variation.…”

Clausius Normalized Field-Based Shape-Independent Motion Segmentation

“…To show the quantitative results, we tested the resulting images using MeanShift (MS) [26], CamShift (CS) [27], and ABCShift (AS) [28] object trackers since we do not have ground truth data. A tracked object is modeled as having a class conditional color distribution, P(C|O), for each pixel with colorC, which is the probability of the color of the pixel, given that the pixel belongs to the tracked object, O.…”