Research On The Detection Technology Of Building Night Scene Lighting Fault Based On Video Surveillance

With the rapid development of urbanization,people are paying more and more attention to the nightlife environment.Building night lighting can not only provide safe lighting functions for urban residents,but also integrate with art,use lighting as a medium to disseminate information,enrich the city’s night scene culture,and improve the overall quality of the city.With the aging of the circuit,the unstable voltage and current,and the outdoor environment and other factors,the lighting of the night scene of the building malfunctions.The existing detection is mainly through circuit fault detection.This method has good real-time performance,but the wiring is complicated,and each light strip cannot be monitored.In response to this problem,the thesis proposes a method of building night lighting fault detection based on video monitoring,which uses outdoor surveillance cameras to collect video images for preprocessing,including image defogging and image registration.Then extract the pixel information from the video and input it into the fault detection model based on the sparse autoencoder for fault detection.The main work of the paper is as follows:(1)Propose an improved dark channel prior defogging algorithm(Improved Dark Channel Prior,IDCP).Due to changes in outdoor weather,the quality of video images will be degraded,especially haze weather will cause the accuracy of the detection of lighting faults in building night scenes to decrease.First,the transmittance is calculated according to the dark channel prior algorithm,the atmospheric light intensity is optimized by the method of removing the maximum value and averaging,and then the transmittance is optimized by the guided filter,and finally the haze image restoration is realized according to the atmospheric scattering model.The test shows that the image after defogging with IDCP algorithm has higher definition and contrast.(2)Aiming at the slight jitter of outdoor cameras,an image registration algorithm based on improved SIFT(Improved Scale-invariant feature transform,ISIFT)is proposed.First,according to the ISIFT algorithm,feature extraction is performed on the night scene lighting image of the building.Then the Euclidean distance and Random Sample Consensus(RANSAC)are used to match the reference image with the image to be registered.Finally,image registration is performed according to the transformation model calculated by RANSAC.The test shows that the average time loss of the ISIFT-based image registration algorithm is 0.062 seconds/frame,which is 24.39%lower than the time loss of the SIFT-based image registration.(3)Designed the detection algorithm of building night scene lighting fault.First,preprocess the extracted continuous frames of the surveillance video to reduce the impact of haze and camera shake.Then,a fault detection model based on sparse autoencoder is constructed,and the pixel information is extracted from the pre-processed video as model input,and the sparse autoencoder is used to enhance the sparsity of features and improve the representativeness and robustness of features.Finally,the K-means algorithm is used to cluster the reconstruction errors,calculate anomaly scores,and judge whether the lights are malfunctioning.Tests show that the accuracy of the lighting fault detection is about 81.28%.