Research And Application Of Image Enhancement Technology For Video Surveillance

Nowadays,with the unprecedented development of network,image processing and microelectronics technology,video surveillance has been widely used in surveillance system because of its simplicity,intuition and digitalization.The quality of surveillance video images will directly influence the performance of surveillance system.The images collected by night video surveillance system are always of low quality in the bad light,which makes it difficult to identify suspicious persons in surveillance video.Therefore,it is a meaningful work to the study of the image enhancement of night surveillance video.Based on the existing image enhancement technology,this paper studies the image enhancement technology according to Retinex theory to prevent excessive enhancement and excessive noise,and combines the night pedestrian detection model to detect only pedestrian images for image enhancement,which reducing the number of images needed to be enhanced in surveillance video.Finally,a surveillance video image enhancement system for enterprise hazard sources is presented based on pedestrian detection model and image enhancement model.Overall,the main work of this paper is as follows:(1)The pedestrian detection model of nighttime surveillance video is studied.Based on the improved Faster R-CNN model,the anchor box can be improved to reflect the shape of pedestrians more effectively.The model can accurately identify pedestrians in nighttime surveillance video and record pedestrian images effectively.(2)An image enhancement model based on Retinex with image denoising is studied.The color ful image containing pedestrians can be transformed from RGB space to HSV space,and only the V(brightness)of HSV space can be enhanced.Aiming at the common problems of over-enhancement and noise amplification in traditional image enhancement,Retinex principle and WVM are used to decompose V,and a series of processes such as shadow lifting and image denoising are carried out on the decomposed illumination layer.This model can enhance the image while avoiding over-enhancement and image noise amplification.(3)A prototype system of video surveillance image enhancement is designed and implemented,and the validity and practicability of the model in engineering are verified.The system contains three main functional modules: pedestrian detection,image enhancement and alarm,which can effectively meet the needs of daily monitoring.The real-world data is used to test the performance of the system.The results show that the system can accurately identify the pedestrians in nighttime surveillance video and enhance the surveillance image at the same time.In summary,the main work of this paper is to study a Retinex image enhancement model,which is combined with the improved Faster R-CNN nighttime pedestrian detection model.The pedestrian surveillance image is enhanced,which further reduces the number of images needed to be processed and improves the efficiency.Finally,the feasibility of the model is validated by designing and implementing a prototype system for video image enhancement.