Semi-automatic generation of accurate ground truth data in video sequences

Abstract

Generation of ground truth data from video sequences is still an intriguing problem in the Computer Vision community. The massive amount of data and the necessary effort for annotating this data make this task a challenging problem. In this paper we investigate the possibility of generating ground truth data in a semi-automatic way. Specifically, using the output of different algorithms, a new output based on robust statistics is generated. The proposed method uses results obtained from real data which is used for evaluation purposes. The generated output is proposed to be used as a basis of ground truth data reducing the necessary time for generating this data. The main contribution of this paper is to show that such methodology can be used to generate an initial ground truth data, which is accurate and reliable, in both ways semi-automatic and fast. Various results and analysis are presented to evaluate the performance of the proposed methodology. Obtained results suggest that generating ground truth data based on the output of different algorithms is possible alleviating the problem of annotating such data manually.