Research On Key Technologies Of Intelligent Object Detection And Target Tracking

In recent years,computer vision technology has achieved rapid development.The technologies of object detection and target tracking have been gradually applied to various fields such as education,medical treatment,transportation,military defense,etc.Therefore,further research on the key technologies of object detection and target tracking has important theoretical research value and engineering application value.With the development of deep learning technology,many research results have been made in the field of object detection based on deep learning.For resource-poor embedded devices,a more lightweight detection model can meet the demand of real-time.Therefore,this paper studies and compares the comprehensive performance of various detection algorithms.Considering the lightweight,real-time,portability and other factors of the algorithm,the researcher of the paper proposed an improved lightweight real-time LR_Det object detection algorithm,which reduces the model storage to 1.3MB,and the speed is 59% faster than YOLOv3-tiny at the same input.In the training phase,this article introduces the CompleteIo U regression loss function to simultaneously train the distance,overlap rate,and scale difference between the prediction box and the real box,which solves the problem of unstable and extremely divergent training of the target box.This method not only speeds up the convergence of the network,but also further improves the performance of the object detection algorithm.In addition,the Batch Normalization layer is merged into the convolution layer during the forward operation.In this way,the time of forward operation of the whole network is reduced by 6%,and the detection speed is further improved.Finally,the superiority of the LR_Det lightweight real-time detection algorithm was verified by experimental results.In the target tracking algorithm,the target tracking algorithm based on correlation filtering can better balance the tracking effect and speed,so research on this type of algorithm has great engineering practical value.This paper first studied the principle and advantages and disadvantages of KCF algorithm and proposed a tracking algorithm based on multi-feature fusion,namely MF-KCF,which combines FHOG features,CN features,and gray features.And the accuracy and robustness of the tracker are improved.In order to solve the problem that the tracker will drift when the target is occluded,deformed,blurred,etc.,this paper uses a high-reliability model update strategy by introducing confidence judgment indicators.Updating the model only when the tracking results are reliable can greatly reduce the probability of the tracking model being polluted and improve the anti-interference performance of the tracker and the tracking speed.This algorithm is called AM-KCF.The high confidence model update strategy will cause the algorithm's learning ability to lag and the tracker cannot adapt to the rapid change of the target when the target appearance information changes relatively quickly.Therefore,this paper combines tracking algorithm and detection algorithm to obtain an effective long-term target tracking algorithm,which can suppress the accumulation of tracking errors to a certain extent,and also has a good description of the latest state of the target.The experimental results show that under the three evaluation methods of OPE,SRE,and TRE,MF-KCF and AM-KCF have improved the accuracy rate by an average of 5.3 percentage points,the average success rate has increased by 10.6%,and the frame rate has remained at 80 FPS,indicating that the improved tracking algorithm is superior to the original algorithm in both accuracy and robustness.