Visual Tracking Based On Reliability In Edge Computing Environment

As a method of secondary verification,reliability has been widely used in various fields such as computer,medical care,transportation and so on.At present,though the development of computer vision in industrial applications is very rapid,there are many problems waiting for solution.Therefore,this paper constructed a reliability measure based on the response graph of region in target matching ? the basic problem in computer vision.The proposed reliability was a secondary verification of target matching to increase the successful matching ratio.First,by extracting the reliability-related features in response graph of visual tracking,an analysis model of reliability was constructed.Positive and negative samples were collected from a large number of videos by aiming at reliability of target matching;and then,a neural network was constructed to estimate reliability by features in the extracted samples.The reliability of matching target position was evaluated by inputting response graph in the proposed model,which was used as assisted decisionmaking strategy for the target matching strategy.Secondly,this paper further constructed the decision logic based on the proposed assisted decision-making strategy,and proposed a novel visual tracking method.This method adopted different template update strategies for different degrees of reliability,and improved tracking accuracy by solving a series of tracking failure under complex environment.Then,this paper deployed the proposed visual tracking algorithm on edge computing environment,and provided an edge computing based visual tracking algorithm.This novel algorithm provided a dynamical joint and release mechanism of computing nodes based on reliability.The mechanism improved tracking accuracy by used in multi-nodes to multi-channel process the multiple suspected targets parallelly.Finally,the proposed method was experimented on the public OTB(Object Tracking Benchmark).In our experiment,(1)A reliability neural network model with a correct rate of 90.9% was obtained,which verified feasibility of the reliability assistance strategy;(2)The tracking algorithm with combination of the model was evaluated to verify the effectiveness of the proposed model.The experimental results showed that OPE(One Pass Evaluation)of the proposed method was increased to 70.8% than classical method,especially,the proposed method was increased 3.38%,2.58% and 2.38% in the fast moving,motion blur background and background clutter by using OTB100 dataset,respectively;(3)The proposed tracking method was further deployed on an edge computing environment.Experimental results showed that the performance of proposed method was further increased with similar tracking speed.OPE of the edge computing-based method was increased to 71.3%.In conclusion,all experimental results showed that the proposed reliability-based method increased the effectiveness of current visual tracking method.