Real-time edge-enhanced dynamic correlation and predictive open-loop car-following control for robust tracking

Robust Edge-Enhanced Fragment Based Normalized Correlation Tracking in Cluttered and Occluded Imagery

“…However, it has been reported in [5] that normalized correlation (NC) is more robust than NCC, when the images to be correlated are edge-enhanced images. Therefore, we use the latter technique.…”

“…Therefore, we use the latter technique. The edge-enhancement procedure comprises: (1) RGB to gray level conversion as in [9] for reducing computational cost without significantly affecting the tracking performance, (2) Gaussian smoothing with adaptive standard deviation parameter as in [5] for attenuating noise without introducing noticeable blur, (3) gradient magnitude computation using Sobel edge detector masks in x and y directions as in [5], and (4) normalization to stretch the pixel values in the gradient images in the whole range of 0 to 255 as in [5], so the object may stand out in low contrast imagery.…”

“…As a result of conventional normalized correlation, a correlation surface is developed which provides matching values between template t, and search image s, when the template is placed at every pixel of search image as [5]:…”

“…EENC [5] has been strengthened by the use of Kalman predictor [6]. When the process of normalized correlation produces the peak value below τ t , the target coordinates estimated by it are disregarded the target coordinates predicted by the Kalman filter in the previous iteration are utilized in the current iteration, and the process of template updating is bypassed.…”

“…Edge Enhanced Normalized Correlation (EENC) tracker [5] has significantly solved the real-world problems of orientation, illumination, obscuration, intermittent occlusion, complex object motion, object fading, and noise. In EENC, the template and the search image are edge-enhanced before performing normalized correlation between them.…”