An Improved CAMShift Algorithm for Object Detection and Extraction

Abstract

Continuously Adaptive MeanShift (CAMShift) is an important algorithm for object tracking based on the colour histogram. The algorithm works by finding the mode of a probability distribution map within a search window and iteratively updates the position and size of the window until convergence. The algorithm boasts of high performance in a simple environment where the colour distribution is constant. However, because the algorithm is dependent on a static colour distribution, its performance suffers in cases where the distribution changes e.g. due to illumination or weather conditions. In addition, object occlusion and complex background colour can degrade the performance of the algorithm. In this paper, we propose a CAMShift algorithm that can track coloured signs. Since multiple colours are involved for tracking, we utilized a Bayesian approach to estimate the colour probability density function. This probability density function gives the probability of whether a pixel value corresponds to certain object. We illustrate the effectiveness of our approach by detecting and extracting visual sign images with different colour attributes. The result obtained shows that our extended CAMShift algorithm can detect and track coloured signs based on the identified colour class.