Object Detection In Moving Camera

With the rapid development of the semiconductor and computer technology, computer vision and image processing has been gradually applied to various fields, such as safe city construction which related with people's daily lives, a large number of intelligent monitoring equipments, Unmanned reconnaissance aircraft and so on. Moving object detection is a key technology in such applications. We propose a motion saliency detection technique for detecting moving object in dynamic scene and put forward a method solving missing objects in the detection, which can suppress unrelated background information when the camera moves in translation.This paper reviews the existing target detection algorithms. We classify the existing target detection algorithms and analyze their principles, advantages and disadvantages. To overcome the shortcomings of these existing algorithms, we put forward a new method for target detection based on dynamic scene:We analysis the motion characteristics of background and foreground with moving camera, we calculate scene saliency map for detecting moving object by simulating the human attention mechanism. The features of moving objects are extracted by optical flow method. In order to suppress background motion texture, we use 2-D Gaussian by convolution and compute the global saliency of the video by counting the histogram. According to the motion saliency, we extract the color information of foreground and background. Finally, dominant moving objects are detected using Bayesian method deals based on motion saliency maps.The algorithm relies on optical flow feature, so objects in some frames may be missed because of incorrect motion estimation when the speed of target is very slow. In order to solve this problem, we address a method by analyzing problem of target lost from perspective of the theory. We propagate the position information of the target based on the optical flow over the entire video sequence; energy map is computed by combing position information and the saliency map in each frame.Experimental results on three public video datasets show that the proposed method can suppress motion texture, while highlight the moving object in the dynamic scene with moving camera, the object position information algorithm can reduce the missing rate of moving object detection.