Target Object Detection using Homographic Transformation

Abstract

Background/Objectives: It is important to detect target objects in complicated environments without particular limits. This paper proposes a new method of detecting a target object in various circumstances with reflection. Methods/Statistical Analysis: The proposed method uses a stereo camera to capture a target object and then extracts the characteristics of the lines and corners that present the object well. After that, homographic transformation is applied to effectively remove unreal reflected characteristics from the extracted characteristics of the captured left and right images. Lastly, the real characteristics without reflected ones are clustered, and a target object is detected robustly. Findings: The test results of this paper reveal that the proposed algorithm detected a target object in the natural environment with reflection more accurately than an existing algorithm. For comparative evaluation of the performance of the proposed object detection algorithm, this paper uses as an experimental image the multiple scenes captured by a stereo camera in the general circumstance that has much reflection and no constraints. In particular, as an experimental image, the inside of an elevator comprised of steel walls with much mirror reflection was captured. This paper clustered real features and split them in the unit of object. To measure the performance of the target detection algorithm proposed in this paper, accuracy measure was used. As shown in the performance evaluation, the conventional algorithm tries to detect a target object without any removal of reflected features so that it has relatively lower accuracy than the proposed method. The proposed algorithm tries to detect a target object after removing unreal reflected features so as to detect a target object relatively reliably. Improvements/Applications: We expect that the proposed technique of detecting target objects in reflection environments will be utilized in various types of real application areas related to multimedia contents.