Research On Intelligent Monitoring System For Workshop Based On Machine Vision

In recent years,with the gradual progress of society,manufacturing safety issues have become more and more important for enterprises.Monitoring safety manufacturing process through video surveillance system is the most common way,but the cost of manual video surveillance is too high and the effect is not ideal.Therefore,intelligent video surveillance systems are becoming more and more important.The main work of the research is to focus on the intelligent video surveillance system of the workshop.In-depth study of its important components:target tracking algorithm and target detection algorithm,introduces the theoretical basis of target tracking algorithm and target detection algorithm,and focuses on the target tracking algorithm and target detection algorithm.The defects were improved.In terms of target detection algorithms,for the purpose of developing target detection algorithms with higher accuracy,lower missed detection rates,and better real-time performance,the CornerNet algorithm is improved,and the basic network of the CornerNet algorithm is improved,and the convolution kernel is used for more Small multiple convolutions replace the larger single convolution of the convolution kernel,and replace the original residual module with a new residual module,which reduces the parameters in the network model,reduces the redundant multi-layer network,and improves The real-time detection makes the detection speed up to 24 frames per second.At the same time,the image pyramid technology is used to expand the training data set,and small-sized training samples are added to reduce the algorithm’s missed detection rate of small targets.The final experiment is divided into two steps.First,the algorithm is quantitatively evaluated through the COCO database,a commonly used database for target detection.The results show that the improved algorithm is better than the existing algorithm in terms of accuracy and real-time performance,and meets the actual needs of intelligent monitoring..At the same time,it also evaluated the surveillance video of the workshop site.The final result showed that the improved algorithm completed the target detection task in real time.In terms of target tracking algorithm,the KCF(Kernel Correlation Filter)algorithm can not complete the tracking task in the face of target deformation,lighting changes and motion blur,and the KCF algorithm has been improved.First,a pre-detection mechanism based on deep learning is introduced.The pre-detection mechanism can obtain more accurate target tracking frame characteristics by using deep learning.Before performing relevant calculations to determine the target position,pre-detect the image to obtain accurate tracking of possible targets.Frame,and then obtain the target position through related calculations to complete the tracking.In addition,the update mechanism of the correlation filter template has been improved,and the linear update mechanism of the residual network model instead of the original correlation filter template is designed.The current image tracking is completed by combining the initial template generated from the initial image and the historical template accumulated during the tracking process.The new template generated later is used as the input,and the output is the updated correlation filter template with better robustness,which replaces the traditional linear update correlation filter template,which improves the robustness of the KCF algorithm.The algorithm is evaluated through the universal target tracking database—OTB100.The quantitative results show that the improved algorithm has higher accuracy,and the workshop site monitoring video verifies the feasibility of the algorithm in the actual manufacturing environment.