A novel multiple moving objects recognition and segmentation based on dense optical flow and K-means clustering

Moving objects detection has always been a prominent research topic for the computer vision and image processing field. In recent years, many methods of moving objects detection in optical imaging have been proposed. However, the high-precision detection and recognition of multiple moving objects is still quite challenging. Many key problems that restrict the performance of multiple moving objects detection, such as dynamic background, complex environment, different objects classification, and detection accuracy, have not been effectively solved. This study proposes a dense optical flow multiple moving objects recognition and extraction method to overcome these problems. Unlike most moving objects detection methods based on sparse optical flow, we propose a multiple moving objects detection and recognition method which combines an optical flow histogram and K-means clustering analysis based on dense optical flow. First, we utilize an enhanced optical flow calculation and feature threshold processing to segment the moving objects foreground region. Subsequently, according to the histogram of optical flow amplitude and directions, K-means iterative clustering is utilized to cluster the foreground optical flow. Different moving objects can be recognized and segmented using different optical flow amplitudes and direction. The implementation results indicate that the proposed method exhibits more optimal results for both rigid and non-rigid multiple targets with small internal optical flow difference in static and dynamic environments.