Detecting salient motion by accumulating directionally-consistent flow

Abstract

Motion detection can play an important role in many vision tasks. Yet image motion can arise from uninteresting" events as well as interesting ones. In this paper, salient motion is de ned as motion that is likely to result from a typical surveillance t a r get e.g., a person or vehicle traveling with a sense of direction through a scene as opposed to other distracting motions e.g., the scintillation of specularities on water, the oscillation of vegetation in the wind. We propose an algorithm for detecting this salient motion that is based on intermediate-stage vision integration of optical ow. Empirical results are presented that illustrate the applicability of the proposed methods to real-world video. Unlike many motion detection schemes, no knowledge about expected object size or shape i s necessary for rejecting the distracting motion.

Into the woods: visual surveillance of noncooperative and camouflaged targets in complex outdoor settings

“…The confidence measure combines the overall target size, its speed in 3D, the quality of the match, occlusion time, its rate of growth (for handling complex lighting false alarms), and its cumulative distance traveled (for handling objects like moving branches before the secondary background model can adapt to include them). The use of cumulative distance traveled is similar in spirit to ideas in [17] though the implementation is significantly different as we do not compute a detailed flow.…”

“…Many use correlation or sum-of-squared-differences (SSD) over windows [17]. These will not work well with the small targets, large amounts of occlusion and target deformations.…”

Joint optimization of background subtraction and object detection for night surveillance

“…The traditional pixel level methods like [1,4] model the background as a set of independent pixel processes, which lose the spatial context information and often end up with noisy detection. Therefore many methods are proposed to utilize the spatial information between pixels [2,3], or to utilize temporal information [5] to better model the background in a scene, or combine both methods [6]. However, in a night outdoor scene, the current existing methods still suffer much from the following problems: 1) heavy false alarm due to dramatic lighting changes and reflections on other static objects; 2) missing detection due to the condition that the foreground is very similar to the background in local due to low contrast.…”

Background Subtraction on Embedded Hardware

“…Discontinuities in the optical flow can help in segmenting images into regions that correspond to different objects. In [180], temporal consistency of optical flow over a narrow time window is estimated; areas with temporally-consistent optical flow are deemed to represent moving objects and those exhibiting temporal randomness are assigned to the background.…”