Abstract:

Photo-cuts are typically display devices placed on a fixed point in the forest land and used to monitor natural life. Millions of images are recorded using photo-covers to investigate the natural life of animals. On recorded images computer-based methods and automatic methods are developed for the purpose of identifying and recognizing animals. Also, problems such as background complexity in photo-cover images, the background being moving, the light strength change and the object being fragmented make it difficult to detect the moving object. In literature for this purpose, model images of moving objects are used as preliminary information in classification-based methods, which are hand-detected from the image. Identifying and scratching the model images of objects by hand is a difficult, difficult, time-consuming process and requires a high workload. In our study, in order to reduce this workload, the photographic-cover images obtained from the natural environment have automatically detected moving objects without the use of the pre-information of objects and the positions of moving objects in the image have been determined. In the suggested method, background-making and frame differentiation methods were applied to the images to identify moving objects. For the creation of the background model, the Gauss Medium and Gauss Mix, the Gauss Extinction and Medium Filter, for the purpose of reducing noise and highlighting objects, have been used OTSU equalization to fix errors in the front plan detection. The success of detection of moving objects in the photo-cover data sets is 83% and the success of positioning objects is 80%.

Keywords:

Abstract:

Camera-traps are usually placed on a fixed point in a forest land and are used to monitor natural life. Millions of images are recorded to investigate the natural life of living things by using camera-traps. Computer based automatic methods are developed for detecting and identifying living things on recorded images. Also problems such as background complexity, moving background, change of light intensity and fragmentations of the object in camera-trap images make moving object detection difficult. In the literature, for this purpose the model images of moving objects obtained from manually are used as preliminary information in classification based methods. Detecting and cropping model images of the objects manually is a difficult, laborious, time-consuming process and requires high workload. In our study, To reduce this workload it was aimed to detect moving objects automatically and to determine the location of moving objects in camera-trap images that obtained from natural environment. In the proposed method for this purpose, background extraction and frame difference methods were applied to the images. Gaussian Average and Mixture of Gaussian were used to create a background model. Gaussian Blur and Median Filter were used to reduce noise and to clarify objects. .OTSU thresholding was used to eliminate the errors of foregrounds. In the camera-trap data sets, the success of detecting moving objects was 82% and the object localization success was 80%.