Moving Objects Detection And Tracking In Low Illumination Conditions

Moving objects detection and tracking is one of the main research contents in the field of computer vision. However, moving targets detection and tracking have been studied rarely in low illumination conditions at home and abroad. But it is gradually getting people's attention because of its widely applied value in all-weather real-time monitoring, traffic control, military and other fields, and it has become a new hot research issue in computer vision field.This paper focuses on video sequence under indoor low illumination conditions with a fixed camera. It includes video sequence preprocessing, enhancement processing, moving targets detection and tracking.For video sequence preprocessing, the reason that signal to noise ratio is low and image noise characteristics in low illumination conditions are presented in this paper., The basic principles of several filters which are commonly used to noise filtering are stated. Then, a spatio-temporal filtering method is used to remove image noise, which Gaussian filtering is used in spatial domain and real-time adaptive inter-frame filtering is employed in time domain. The experimental results show that the proposed method is effective.The high-frequency noise is also amplified when Logarithmic processing algorithm is used to enhance the low illumination video image. A multi-level Logarithmic processing algorithm is used to overcome this shortage, and its parameters are used to control high-frequency sharpening components of each level image simultaneously. As a result, the amplification of the high-frequency noise is suppressed effectively. In order to expand the whole grayscale dynamic range of enhanced image, an improved contrast stretching algorithm is used to enhance video image contrast. Experimental results show that the proposed approach achieves a good enhancement results.In low illumination conditions, it is not very accurate to update the background merely on the base of the current frame mask, which leads to misjudgment of the background point when the Gaussian model method is used to build background model. Therefore, an improved the update algorithm of Gaussian background modeling based on the overall effect of the masks of successive three frames is adopted, which can update the background timely and accurately. Experimental results illustrate that the detected moving objects are more complete and accurate. In order to improve the accuracy of tracking moving objects, a new approach is adopted, which determines whether occlusion occurs among the moving objects based on the number of the predicted objects in the detected moving region. Before the occlusion occurs, a matching method based on normalized moment of inertia value is used to look for the subsequence target. When the occlusion occurs, a template matching method is adopted to track the objects based on the Kalman filtering prediction, and the Kalman filtering model is updated based on the best match location. Experimental results show that the algorithm can achieve an better tracking effect.