A Robust Object Tracking Algorithm Based on Surf and Kalman Filter

Abstract

In this paper, an efficient robust object tracking approach based on SURF and Kalman Filter is proposed. SURF as an outstanding local invariant feature is employed. Based on the SURF feature, a SURF match method is proposed. A combination method using an ingenious method and KF is used to predict the possible region, in which the tracking object may appear. Only in this region, SURF features are extracted and matched. It can significantly reduce the computational complexity. A histogram-based re-match process is employed to dislodge failure tracking after SURF matching. To verify the performance of the proposed algorithm, several comparative experiments are conducted. The results reveal that the proposed method achieves better performance and accuracy than conventional methods.

Continuous discrete cubature quadrature Kalman filter

“…In nonlinear filtering, the discrete-time filters are popular and a wide variety of them are available in literature [1][2][3][4][5][6][7][8][9][10][11][12]. It is commonly applicable in many real-life problems like target tracking [1,[13][14][15][16], navigation [17], system identification [18], etc. In discrete-time systems, both the process and measurements are in discrete-time domain while practically many times the process is described in continuous time domain.…”

Robust tracking of moving objects using thermal camera and speeded up robust features descriptor

“…The described approach 26 provided an idea for a thermal image tracking algorithm design, by combining a SURF algorithm and Kalman filter. An algorithm that uses SURF features and Kalman filter to track objects in color imagery is presented in Reference 24, where a standard Kalman filter is used to predict the position of a bounding box around the object in the next step (where the SURF features will be searched in the next step), in order to decrease processing time. If errors occur (due to occlusion or improper matching), the problem is solved by a histogram rematch algorithm.…”

“…If errors occur (due to occlusion or improper matching), the problem is solved by a histogram rematch algorithm. That approach, as described in Reference 24, provides good results, but since the histogram rematch algorithm is an additional algorithm, it increases processing time. Described in Reference 13 is an algorithm that combines several algorithms (CamShift, SURF, and Optical Flow) and Kalman filter matrices are tuned depending on the algorithm error.…”

“…Main problems with these kinds of trackers is that they are hand crafted for specialized problems, therefore do not show robustness in challenging situations. Algorithms that use object descriptors, such as SIFT 11,23 or SURF 13,[24][25][26][27] are very popular in object tracking because they are robust to partial occlusions, scaling, and rotation. 28 However, not all descriptors can be employed with the same efficiency in both visual and thermal domains because the characteristics of the images differ considerably.…”

Optimized object tracking technique using Kalman filter

“…Tracking objects in the real time environment is not a trivial task and has been a popular research topic in the computer vision field. A robust vision tracking system allows for many applications such as traffic monitoring, autonomous system guidance, surveillance systems, human computer interaction, vehicle navigation, and automated weapons systems [2,4,5].…”