Wide Array Design Of A Distributed Optical Target Detection And Location System

Forest fire problem has gradually become a global problem, causing huge economic losses and environmental damage. Currently, the traditional forest fire monitoring system is mainly based on a single camera. The detection range of a single camera is limited by the manpower to monitor, which can not cover the whole forest. The practical application of intelligent monitoring fire can`t occur.Distributed array camera design approach has great advantages in the wide space of group target detection. Fan-shaped arrays composed of multiple cameras to cover the entire forest range while the camera group consisting of multiple arrays can pinpoint locations of fire. The features of a single camera image, using support vector machine classifier, build a fire recognition model and intelligent monitoring of fire. Singlecamera surveillance overcome the flaws of human monitoring. The final design complete array of wide-area distributed optical target detection and location system. The main work is as follows:1. Analyzes the key technology which multi-camera array and target detection image stitching involved. On this basis, for some algorithm we have done a detailed analysis.2. Study the different characteristics of the image, and detect the effect of a variety of features, focusing on comparing different color features described in the performance of the flame through experiments which confirmed the feasibility of color features in the flame detection.3. We studied the LBP(Local Binary Pattern) feature, trained by a flame detector classifier LBP+SVM(support vector machine) method. And we propose to combine LTP(local ternary mode) and SVM for the use of flame detection which overcomes the shortcomings of LBP feature in dramatic light or noise of extreme imaging conditions, and SVM classifier performance degradation. Then LTP improves the robustness of the flame detection. Compared to the LBP features, detection results of classifier based LTP features is better.4. Based on the LTP+SVM classification detection model we design the front part of the whole system. The front part of the whole system is obtained by image stitching module extending the entire FOV. The target is detected by the flame detection module LTP + SVM classification. After the flame strike azimuth module the azimuth is obtained and transferred to the back-end. The front accepts the back-end rotation parameters, and controls pan.5. Software for the back-end system is implemented based on OpenCV and MFC. Design time domain filtering module, which within a specific time is considered false detection when Azimuth of single array is returned, for effectively preventing false detection from the front. Eventually through two or three-way azimuth we get GPS coordinates of fires through the intersection module, and the software alerts, while the rotation parameters which is computed by the GPS coordinates returns to the front-end system. Then the software accept front head signal and displays the video stream.