Stably Adaptive Anti-Occlusion Siamese Region Proposal Network for Real-Time Object Tracking

Abstract

Siamese region proposal network has made remarkable achievements in visual object tracking because of its balanced accuracy and speed. However, it regards tracking as a local one-shot detection task, which lose the power of updating the appearance model online thereby cannot handle the object-occlusion, fast motion and out-of-view situations. To tackle this problem, we propose a method that combines adaptive Kalman filter with Siamese region proposal network (Anti-occlusion-SiamRPN) to make full use of the object spatial-temporal information. Specifically we first extract target features through deep network and then uses adaptive Kalman filter to predict target trajectory in these difficult scenarios. Further this trajectory is used to select the candidate area of the next frame for Siamese region proposal network, which improve the searching mechanism. In this way, the introduction of adaptive Kalman filter makes the tracking process online learning which makes up for the disadvantage that Siamese region proposal network can only track offline. In addition, a hard example discrimination method (HEDM) is proposed to estimate whether the occlusion occurs and how seriously it is, which also improve Kalman filtering mechanism to make it update adaptively. Our method being evaluated with the speed of 80 FPS on five widely-applied challenging benchmarks including OTB2013, OTB2015, OTB50, VOT2016 and VOT2018. The extensive experimental results demonstrate our method achieves state-of-the-art effects and great improvement in comparison to other trackers. INDEX TERMS Visual tracking, Siamese region proposal network, Kalman filter, occlusion judgment, real-time tracker.

Siamese Visual Object Tracking: A Survey

“…In Table 1, we provided a recent quantitative comparison of surveyed trackers on established benchmarks. We can see that the top-performing trackers are based on region proposals [61], [63], [69], [86]. The comprehensive survey from Marvasti-Zadeh et al [17] also reached a similar conclusion.…”

“…Considering the OTB 2013 scores, we see that the best performing tracker is SE-SiamFC. However, we also have to emphasize very similar performance of FIGSiam[56], SA-Siam[16], and AO-SiamRPN[86]. There are two reasons.VOLUME 4, 2016 …”

“…The robust performance of the SiamRPN architecture inspired multiple subsequent works. One of them was proposed by Wu et al [86]. The authors developed a modified SiamRPN architecture with an anti-occlusion mechanism based on Kalman filter [87] dubbed as Anti-occlusion Siamese region proposal network (AO-SiamRPN).…”

“…Values marked with a superscript number were obtained from other than the original paper. Alternative sources of data were: a [55], b[41], c[16], d[98], e[63], f[40], g[57], h[86], i[56], j[69], k[20], l[106].…”

GOA-Net: Generic Occlusion Aware Networks for Visual Tracking

“…Deepak et al [17], in their work SiamFC-SD, used structured dropouts in feature maps to mimic the changes under occlusion. Wu et al [18] used a hard example discrimination method to estimate occlusion occurrence. Wenil et al [19] used depth information to predict occlusion and precise object location.…”