이동 물체 검출을 위한 향상된 그래디언트 기반 MKFC 알고리즘

초록

For moving objects detection, the multi-channel kernel fuzzy correlogram algorithm(MKFC) has the following two advantages. It detects moving objects not only sharply by a pixel-based modeling algorithm, but also robustly against dynamic background by a region-based modeling algorithm. However, it uses a quantized pixel value to reduce computational complexity. For this reason, it is hard to apply the gradient feature (e.g., local binary pattern(LBP) and histogram of oriented gradient(HOG)) which is robust against the light change environment. To solve this problem, we propose an algorithm that detects the moving objects using gradient vector-based MKFC. However, performance of moving object detection can be degraded by the noise resulting from gradient vector-based image transform. In this paper, we propose an improved moving objects detection algorithm using gradient based MKFC. The algorithm uses a median filter for denoising the image from gradient vector-based image transform and determines the candidate region based on similarity estimate. Finally, the approach detects the moving objects by applying a Gaussian filter to the candidate region. Simulation results in a light changing environment show that the proposed algorithm yields better moving-object detecting performance than the conventional algorithm.

키워드