Unsupervised moving object segmentation and recognition using clustering and a neural network
Jong Bae Kim, Hye Sun Park, Min Ho Park, Hang Joon Kim · 2003
This paper presents unsupervised moving object segmentation and recognition method for intelligent transportation systems (ITS). The presented method consists of three procedures: First, the object detection procedure, in which the rough positions of moving objects in an image sequence are determined using an adaptive thresholding method; Second, the object segmentation procedure, in which pixels that have similar intensity and motion information segments are grouped using a weighted k-means clustering algorithm to the binary motion mask obtained in the object detection. Finally, the object recognition procedure, in which a neural network is used to recognize whether the segmented objects are vehicles, humans, or other objects. The experimental results demonstrate robustness not only in variations of luminance conditions, but also for occlusions among multiple moving objects.