Research On Video Synopsis Method Based On Object Trajectory Rearrangement

Video Synopsis aims at eliminating the redundant information of surveillance video and generating condensed video for fast browsing.Due to the characteristics of preserving the dynamic information of moving objects,video synopsis is widely used in applications such as video investigation and suspected object retrieval,and becomes a hot issue in the field of computer vision and video processing recently.However,traditional synopsis methods are not dealing well with the contradition between high compression and low collision phenomenon when rearranging the tubes,which is the core problem in dynamic video synopsis.To solve it,this dissertation focuses on rearranging tubes to the optimal temporal location in synopsis with different strategies,such as judgement strategy,graph model,stepwise processing framework,and objects classification,etc.Firstly,we propose an enhanced synopsis method by incorporating judgement strategy into the tube rearrangement model.It turns out that the proposed method causes fewer collisions and preserves chronological order between tubes with less computational complexity.Experimental results on several surveillance videos demonstrate the effectiveness of the method in efficiency,temporal stability and collision reduction.Secondly,to deal with the problem of spatial-temporal contradiction between collision and compression,we propose a novel tube rearrangement approach based on graph coloring.The input tube set is represented by a graph structure,where each node stands for a tube and the edge between two nodes represents the potential collision relationship.Then the tube rearrangement is formulated as a graph coloring problem.To mitigate the collision artifacts,our method finds the mapping of tubes from original video to synopsis video by coloring the graph,which separates tubes from their collision points.Moreover,the coloring interval is left tunable to make a compromise between collision artifacts and synopsis length,which can better meet users’ demand.Extensive experimental results show that the proposed method can generate more compact video synopsis with less collision artifacts than the existing methods.Thirdly,to improve the speed of synopsis generation,we propose online video synopsis method based on potential collision graph.A potential collision graph is constructed to represent the tubes and their potential collision relationship.The procedure of graph coloring is changed to directly add constraints on nodes.Moreover,we incorporate tube rearrangement into a stepwise online framework to further improve its efficiency,which formulate the global optimization on tube rearrangement to stepwise optimization problem.Experimental results show that the proposed method can efficiently generate more compact synopsis video with less collision artifacts and temporal chaos.At last,synopsis video may bring uncomfortable visual effect to users when the input video contains lots of objects.To solve it,we propose a video synopsis method by combing pedestrian-vehicle classification.With the object sequences extracted by video synopsis,the pedestrian-vehicle classification can construct the category index of tubes.As a result,users can generate the synopsis video that includes pedestrians only,vehicles only or both on request.Experiments are conducted to evaluate the performance of synopsis generation incorporating classification.It shows that the pedestrian and vehicle are well classified and thus the corresponding synopses on different categories are more compact with less collision artifacts.Moreover,the video synopsis software incorporating classification is user-friendly in communication.