A Robust On-road Moving Platform Video Stabilization Using Derivative Curve Warping Deepika Shukla 1 and Rajib Kumar Jha 2 1 PDPM Indian Institute of Information Technology, Design and Manufacturing Jabalpur, India 2 Indian Institute of Technology Patna, India Abstract. In this paper a video stabilization technique based on deriva- tive of integral projection curve has been proposed. The existing projec- tion matching techniques consider only intensity values as the matching feature. Any change in pixel intensity values and their local distribution, results in shape variation of the projection curve. This shape variation affects the accuracy of curve matching and becomes a serious cause of wrong motion estimation. The proposed derivative curve technique con- siders the change in intensity values as the matching feature, instead of working on the intensity values itself. Inclusion of derivative provides shape information as high level feature and this results in better matching of two slightly dissimilar curves which are obvious for sequences taken from moving platform. Optimal matching of these derivative curves is achieved using Dynamic Time Warping technique. Keywords: Integral projection, Dynamic time warping, Video stabilization. 1 Introduction The videos captured using hand-held or moving platform camera are prone to undesired vibration induced motions. These undesired high frequency motions affect the visual quality of video sequence significantly. Video stabilization aims to remove the effect of such unwanted motions to provide a pleasant viewing ex- perience. The relative motion between two consecutive frame is obtained using some motion estimation techniques. Accuracy of the estimated motion is highly desired as it directly reflects the performance of the whole stabilization system. In literature most of the motion estimation techniques are based on the block matching. The full block matching (BM) [1] technique provides accurate results in case pure translation but get affected by the presence of intensity variation or change in the local intensity distribution. It also suffers from large computa- tional time. Robustness to intensity variation can be achieved using gray coded bit plane matching (GCBPM) instead of intensity frame and working on bit level reduces the processing time also [2]. Hybrid of these techniques have also been proposed to improve the accuracy of the motion estimation by working on a portion of frame to reduce the computational complexity [3]. Reducing the P. Maji et al. (Eds.): PReMI 2013, LNCS 8251, pp. 343–348, 2013. c  Springer-Verlag Berlin Heidelberg 2013