Moving object detection finds lot of application surveillance and traffic monitoring. There are many methods for object detection. Object detection generally involves Key-Frame extraction and Background Subtraction. Key frame extraction summarizes video by eliminating transitional frames, thus reducing computational load. Frame difference, local binary differential, wavelet based and histogram based methods are some of the methodologies used to obtain key frames.
KEY FRAME EXTRACTION
The frames which contain all the important information of a video shot are called the key frames of that video shot. Current day, Video sequences that have high frame rate and high resolution. A video of frame rate 25fps and 1920X1080 grayscale image produces approximately 6MB of data every second. The amount of data to be processed can be reduced by choosing only the key frames from video shot. Key frames are set of frames that summarize a video shot. Applying image processing algorithms only on key frames reduces computational load with little loss in generality. A video sequence containing moving object has many transitional frames that have very minimum displacement when compared to its position in the previous frame.Key Frame Extraction techniques drop frames that do not show significant changes with respect to previous frames while maintain the essential information required for human eye to judge the direction and apparent change in motion of the object. Such techniques find application in video compression, content based video retrieval and video classification.
Most algorithms for key frame extraction involve finding inter frame distance and comparing it with threshold. Wavelet Transformation, local binary pattern (texture based) and key point based are other popular techniques. Wavelet based techniques look at various sub-bands and their relative changes to determine key frames. Texture based techniques divide the images into blocks of fixed size. Operators like first order differential are applied to each block. The result is compared with multiple thresholds to decide keyframes. Key point based, attention based techniques look for certain points in image (key points) that are of significant importance and track them. A large change in such points is needed to classify the frame as key frame. This section surveys such methods.
An algorithm for key frame extraction based on improved optimized frame difference taking different weight, position information of pixels into consideration is proposed in . The given image is divided into 9 regions and assigned weights. The weights are assigned such that center of the image is given more importance than corners. Hence the frame with significant change in the center of the image pops out as keyframe. The metric used is the weighted histogram difference between two frames. The equation is as follows:
where Ht denotes image histograms for the t-th frame and Hkt denotes the values on the k-th grid of...