International Core Journal of Engineering 2020-26 | Page 98

2019 International Conference on Artificial Intelligence and Advanced Manufacturing (AIAM) Moving Target Detection using Inter-Frame Difference Methods Combined with Texture Features and Lab Color Space Fan Gao Yonggang Lu * College of Information Science and Technology Lanzhou University Lanzhou, China [email protected] College of Information Science and Technology Lanzhou University Lanzhou, China [email protected] Abstract—Moving target detection from a video has been considered as one of the most crucial tasks in computer vision. It is originated from the object recognition of still objects. Moving target detection is still far from being completely solved due to complex conditions, such as transformed background and changing light. In this research, some classical moving target detection methods have been summarized, and a new inter-frame difference algorithm for moving target detection is proposed. The proposed method is based on a two- frame-difference method using texture features and color features. The method can be used to detect moving targets under a background with a little perturbation. The experimental results show that combining the texture features and color features can improve the performance in the moving target detection. of candidate region. In the Faster R-CNN method, the first step is to select candidate areas and to obtain a feature image by convolution of several layers of the input image. Then it generates candidate areas on the feature image to extract N ROI from the images to be tested. In the second step, feature extraction is performed. In the third step, based on the features obtained in the second step, classification is done. The last step is location refinement using bounding box regression. When video analysis emerges and the problem of moving target detection appears, computer scientists naturally think of applying machine learning methods to the moving target detection. However, this kind of methods is not suited for the target detection in video very well. First, applying the deep networks to all video frames introduces an unaffordable computational cost. Second, recognition accuracy suffers from deteriorated appearances in videos, such as motion blur, video defocus, etc. However, these problems are seldom observed in still images. Therefore, the deep neural network is not suitable for moving target detection. Keywords—Moving target detection; Inter-frame difference; Gabor Filter; Lab color space. I. I NTRODUCTION Moving target detection in video is a fundamental and critical task in computer vision [1] and its purpose is to detect the moving objects in the video. It is the image processing procedure for extracting moving targets which have apparent movement compared to background. It can be detected basing on its characteristics in frame sequences. Owing to video analysis algorithms are all targeted at the pixel points in the target area, moving target detection is a fundamental pre-processing step in computer vision and video processing, such as visual surveillance (e.g. action recognition), smart environments (e.g. parking occupancy detection and fall detection), and target tracking or classification. B. Detection based on Inter-frame difference Classical algorithms based on Inter-frame difference [4] obtain the moving target contour by differentiating two adjacent frames in video image sequence, as shown in Fig. 1. The absolute value of the difference can be obtained by subtracting the two adjacent frames. They make use of the continuity of video sequence collected by camera. If there is no moving target in the scene, the change of continuous frames is very weak. Inter-frame difference detection has some pros and cons. It works well on situations where there are multiple moving targets and perturbing background, and it has a fast calculation speed. However, it can only extract the boundary of targets and cannot extract the complete area of the targets. Moreover, the selection of inter-frame interval also has a great impact on the result of the target detection, that is, the phenomenon of "double shadow" and "picture cavitation" is easy to appear. Moving target detection [2] has been studied for a long time. There are many kinds of proposed methods, such as CNN based detections [3], inter-frame difference methods [4], optical flow methods [5-6] and background subtraction methods [7]. A. CNN based detection With the development of computer and the rise of machine learning, deep neural network has been widely used in the target detection from still images. For example, a classic algorithm, Faster R-CNN [8], utilizes two Convolutional Neural Networks: one is the region generation network to obtain each candidate region in the image, the other one is the classification and border regression network 978-1-7281-4691-1/19/$31.00 ©2019 IEEE DOI 10.1109/AIAM48774.2019.00022 Fig. 1. The flow chart for the inter-frame difference algorithm. 76