Object Tracking in Occlusion and Contrast Conditions using Patch-wise Sparse Method

摘要

Object detection and tracking is one of the active areas of studies because of variation in object location, pose, illumination, scale and occlusion.Singular object tracking (SOT) is an important task in human-computer interaction, anonymous tracking in the robotic application and security analysis.This paper proposes single visual tracking in a complex environment.We propose SOT using patch sparse formulation that is robust to pose, scale, logterm occlusion and low-and high-luminance conditions.A sparse dictionary using the patches from the previous frames is used to track object in the current frame under such a clutter environment.Contrast modeling, patch resizing and likelihood measurement in the proposed sparse framework allows the selection of key patches to tackle long-term occlusion and extreme illumination variation.We evaluate the proposed method on different challenging datasets involving Illumination variation, heavy occlusion and pose variation.The result analysis indicates that the proposed patch wise joint sparse occlusion and contrast (PJS-OC) tracking method outperforms other methods with the maximum precision rate is 99%.Also for the sampling value of 0.5, the proposed tracker shows excellent results compared with the other different sampling values.The proposed sparse formulation provides excellent performance in various quantitative parameters, achieving better precision and speed.The comparative analysis with state-of-art methods using the benchmark database shows the outstanding performance of the proposed method.