Abstract In recent years, detection and tracking people from a video stream have been widely studied in many commercial and public surveillance systems. Moving object detection is considered as a crucial phase of automatic video surveillance systems. Particularly, people detection is the first important step in any technique of video tracking processes which can be divided into many stations such as... In recent years, detection and tracking people from a video stream have been widely studied in many commercial and public surveillance systems. Moving object detection is considered as a crucial phase of automatic video surveillance systems. Particularly, people detection is the first important step in any technique of video tracking processes which can be divided into many stations such as motion estimation, tracking people, etc. Several methods have been developed for this problem of separating the foreground and background pixels in video surveillance. This paper focuses on computable evaluation of some people detection algorithms for four different video sequences. Our study is based on quantitative and qualitative results respectively by calculating the loss of foreground pixels. In this study, three methods have been evaluated by using two metrics: False Negative Error (FNE) and False Positive Error (FPE). In the result we choose the algorithm which minimizes the Error (%). Practically, the good technique which dominates on the video surveillance applications is the statistical representation of pixels in foreground which is named Gaussian Mixture Model (GMM). In the second part of this paper we control the people entering in a supervised region by detection of the first correct pixel incoming to our supervised area, and we trigger an alarm system. Read More Read Less

Abstract The video summarization is the process to present a rapid view of the important video scenes. It is a necessary step to create an efficient system of indexing and retrieve video. Mainly summarization methods apply a clustering algorithm, which use the dissimilarity matrix calculated between all video frames to extract the key frames which imply a quadratic calculation. To reduce... The video summarization is the process to present a rapid view of the important video scenes. It is a necessary step to create an efficient system of indexing and retrieve video. Mainly summarization methods apply a clustering algorithm, which use the dissimilarity matrix calculated between all video frames to extract the key frames which imply a quadratic calculation. To reduce this complexity, we propose a hierarchical approach to extract video summary based on shot segmentation and motion estimation. In the first step, we use an adaptive detection of shot transitions to segment the video in shots. In the next step, we apply a hierarchical clustering in each detected shots to extract the keyframes; the number of these keyframes is adaptive to the motion in the shot. We propose a statistical approach using a co-occurrence matrix to estimate this motion. To validate the effectiveness of the proposed approach, we present some experiment results based on real video. Read More Read Less

Abstract Nowadays, it is inconvenient for housekeeper parents to constantly watch over their newborn baby while doing their work or chores. This paper proposes a simple voice recognition system which can be applied practically for designing a device with capability to detect a baby’s cry and informing the parents automatically. There are a lot of similar projects and experiments which have... Nowadays, it is inconvenient for housekeeper parents to constantly watch over their newborn baby while doing their work or chores. This paper proposes a simple voice recognition system which can be applied practically for designing a device with capability to detect a baby’s cry and informing the parents automatically. There are a lot of similar projects and experiments which have been performed recently, but most of them are about recognizing and classifying different types of crying (like for hunger, etc.) and have used complex methods of implementation such as neural network. But in this paper our aim is to merely detect infant’s crying, and our solution is to use a fuzzy classifier which is easy to implement and fast to execute. The overall algorithm is to evaluate the resemblance of the infant’s voice signal with the data stored in a database, which is already prepared by recording some cry and laughter samples, using an automatic fuzzy classifier system which can lead to detection of cry or laughter. This algorithm can serve as a reliable foundation on which the future creation of a portable real-time, automatic voice detection device can be based. It is a pretty formidable task to implement complex algorithms, such as neural networks, on common available microcontrollers, however we proposed a much simpler algorithm which enables us to develop a real-time and low cost device. To evaluate the algorithm, we have created a database of sample cry and laughter signals and developed a sample Matlab program for carrying out the real-time frequency-domain calculations and a sample visual program in Labview programming environment for interfacing with user. Read More Read Less

Abstract The notion of super − resolution algorithm is production of a high resolution image from multiple low − resolution images, to success this procedure, we must have an accurate motion estimation of low − resolution images; this latter is a key element of increasing the resolution of an image. In this paper a novel framework for motion estimation is presented,... The notion of super − resolution algorithm is production of a high resolution image from multiple low − resolution images, to success this procedure, we must have an accurate motion estimation of low − resolution images; this latter is a key element of increasing the resolution of an image. In this paper a novel framework for motion estimation is presented, this framework consists to estimate the shift motion of high-resolution image using a sub-pixel registration. In the simulation we need a set of low − resolution images created from a high − resolution image to perform the sub − pixel registration using the phase correlation, to estimate the shift motion between each low-resolution image and the reference image. A high − resolution image is then reconstructed using bicubic interpolation. Finally, this framework is tested and compared to other frequency methods of shift estimation, the simulation results show the performance of our framework than the other frequency methods of super − resolution. Read More Read Less