| The goal of finding the presence and locations of human faces in arbitrary given images is a difficult task,but has gained more attention over the last 20 years.The task is to detect a specific instance of any object for recognition problem in computer vision.Although much research has been done,current methods for detection purpose work well in maintained controlled settings,but in real world conditions,where light work differently is inconsistent,occlusions are common,and object's pose is not in controlled position,the previous methods do not perform well,also object detection and recognition algorithms are computationally expensive and so that's why much of the work done currently in this field is to optimize these algorithms.Face detection is an extremely active subset of object recognition because it has many applications,especially in the security and surveillance field.In our security obsessed world,government bodies are extremely interested in what surveillance tasks can be automated and in turn are investing heavily in this field.These tasks include tracking the movement of people in video feeds using static or moving cameras and facial recognition systems in which face detection is the first step.There are also numerous applications in the consumer realm,for instance auto-focusing on faces in digital cameras,Wii-style gaming,and head-tracking webcams.For the majority of these applications to be successful,detection must happen in real-time.This adds another level of difficulty to an already complicated problem.I am applying a popular face detection method "Robust Real time Face Detection",proposed by Paul Viola and Michel Jones,which is now the current method of choice using in the Open Computer Vision Library(OpenCV).Based on simple features,encodes the basic image properties.The features are individually not very powerful,but can be computed rapidly.When these obtained features are combined together,they can detect the faces accurately.The algorithm also uses machine-learning technique "adaboosting" to construct sets of important features that will quickly detect faces.This method has three key points that makes the detector to achieve high performance boosts over previous systems.The first is a new image representation called the "integral image" for faster and efficient feature computation.The second is the use of the machine learning technique called AdaBoost(adaptive boosting)for the purpose of selecting simple and strong classifiers.The last is a method of combining these computed classifiers into a "cascade" that quickly eliminates background regions and on more promising areas of the image.The high detection rate is achieved on graphic processing unit(GPU)and the results are then compared with the CPU results. |
