Research On Object Detection Technology Of Video Sueveillance For Public Buildings And Its Application

With the increasing emphasis on the security of public buildings and the continuous development of building intelligence,public building safety protection has become an important part of modern intelligent buildings.The traditional method of public building security protection mainly use video surveillance for comprehensive and effective monitoring.But this kind of passive way is obviously unable to meet the security needs of modern large public building groups.The video surveillance is required to have an ability of intelligent monitoring and can actively analyze video pedestrian information.Therefore,the object detection technology has become the focus of widespread attention in today’s society.In modern public building security,video surveillance is required to be able to actively and quickly detect pedestrian targets at night.When abnormal personnel are found,they can proactively issue warnings to remind security personnel to patrol and overcome traditional methods of relying on security personnel.In addition,video surveillance needs to have the function of proactively detecting human faces,and can save these face images after clear processing,which is convenient for cooperating with public security department or related other departments in handling cases.The specific research contents of this paper are as follows:(1)In view of the environment where video surveillances stay at is dark in public buildings and insufficient lighting at night,it is easy to cause the low contrast and brightness of video images.Enhancing the RGB image directly will cause unnecessary information enhancement,which will increase its calculation amount and has poor effect of enhancement.This paper proposes a fused image enhancement method which is based on color space.First,the original video image is converted from RGB color space to HSV color space.Then the histogram equalization and multi-scale Retinex algorithm are fused to enhance the V component.Finally,the image is return to RGB color space and use Gamma correction to further enhance.The experiment has achieved a good enhancement effect on video image,and reduced the amount of calculation,which has laid a foundation for improving the accuracy of subsequent corresponding algorithms.(2)Because the image detection of video surveillance in public building security does not use the information of the pedestrian movement,the detection result is prone to false alarm targets.This paper proposes to combine the GMM modeling method with the YOLOv3 algorithm,and for the real-time requirements of pedestrian detection in practical security application,the YOLOv3 algorithm is improved by lighting model structure and using k-means dimension clustering to determine the appropriate prior box.An improved GMM-YOLOv3 video pedestrian detection algorithm is proposed.Experimental results show that the algorithm further improves the detection speed,and effectively solves the problem of false alarm targets,thereby improving the detection accuracy.(3)As a specific application example of object detection,face detection uses MTCNN cascade network for face detection for faces with different scales in video surveillance under public buildings.On this basis,batch normalization(BN)is added,and a BN-MTCNN network model is established to improve the performance of the model.Then the Kernel Correlation Filter(KCF)algorithm is used to track the detected face position to reduce the missed detection rate.At the same time,the detected face image is processed for clearing.Combined with the idea of style transfer,the detected face image is transferred with the normal high-definition face image.The clear image generated by image reconstruction and gradient constraint using Poisson image editing method.The experiment shows that the algorithm can effectively improve the multi-scale face detection performance,and the clear processing method has an obviously improvement effect on the processing of face images.