OBJECT DETECTION WITHIN IMAGES AND IN VIDEO STREAM BASED ON INTEREST POINTS Amin Mohamed Ahsan, Prof. Dr. Dzulkifli B. Mohamad Faculty of Computer Science and Information Systems, Universiti Teknologi Malaysia am2002as@gmail.com, dzulkifli@utm.my ABSTRACT In computer vision, object detection is an essential process for many other applications such as object tracking, object recognition, objects categorization, event detection, and searching for object in sequence of images or video. It also has a high relation to other fields such as robots control and medical imaging. Detecting an object within images or in video stream involves three steps at least which are features extraction, object classification and object localization; it depends on researcher’s strategy. In this study and as first step, we present a method to extract local features based on interest point which is used to detect key-points within an image, then, computing histogram of gradient (HOG) for the region surround that point. Proposed method used speed-up robust feature (SURF) method as interest point detector and exclude its descriptor. The new descriptor is computed by using HOG method. The proposed method got advantages of both mentioned methods. To evaluate the proposed method, we used well-known dataset which is Caltech101. The initial result is encouraging in spite of using a small data for training; the classifier used to examine features that are obtained by our method is k-nearest neighborhood (k-NN). Detection rate, specificity, and precision which are obtained using our method are, 0.85%, 97.8%, and 90.5% respectively. Currently, we are working on enhancing the classification and localization methods using a huge data. KEYWORD Object Detection, SURF, HOG, k-NN