All-day moving objects detection for security at level crossing

In this paper, we propose a strategy based on the joint use of background/foreground segmentation methods and colorimetric invariants or color spaces, in order to detect more precisely moving objects at level crossings throughout the day. The proposed strategy is composed of three steps : 1/ apply an adapted colorimetric invariant on the acquired image in order to simplify the image and limit the brightness changes recorded throughout the day, 2/ use a common background subtraction algorithm (Codebook) on simplified images, 3/ track moving objects using a Kalman filter in order to visualize the benefit of this approach at the end of treatment. Results obtained illustrate the common use of color invariants with a Codebook-based background subtraction method in order to provide better segmentations results on images that do not correspond to the current learning state (compared to those obtained without the use of colorimetric invariant/color space). To show the effectiveness of this method, a mobile objects tracking is performed on obtained segmentations.