Scholarlink

<p>Vehicle tracking in road scenarios needs to adapt to varying illumination conditions, while also handling challenges such as similar vehicles, different occlusions, and image blur caused by camera shake or fast vehicle motion. Considering the road surveillance pattern and the vehicle’s appearance and motion changes, this paper proposes a Siamese vehicle tracker for road surveillance, namely SiamVTRS. It achieves a pragmatic and efficient combination of SiamCAR, a Kalman filter with camera motion compensation (CMC), and an adaptive template update mechanism. Additionally, SiamCAR is appropriately optimized and equipped with the average peak-to-correlation energy (APCE) assessment, which can perceive and suppress the common distractions in road scenarios. Furthermore, a re-capture strategy considering the target prior’s state is proposed to assist tracking when the target is missing or faces severe distractions. It performs global detection of candidate vehicles based on the target’s prior location and appearance state, aiming to rapidly re-locating the target. Extensive experiments are conducted on the UA-DETRC, DTB70, and UAV123 datasets, where the success rates of SiamVTRS on vehicle sequences reach 79.2%, 88.2%, and 91.8%, respectively. Compared with other mainstream trackers, SiamVTRS achieves competitive performance and demonstrates superior vehicle tracking in diverse road scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SiamVTRS: enhanced Siamese vehicle tracking for road surveillance

  • 期刊论文

  • 发表日期：

    2026年7月27日

查看全文
  • Shusen Guo,
  • Xianwen Yu,
  • Youchen Liu,
  • Dehao Ma,
  • Chenkai Xu

摘要

Vehicle tracking in road scenarios needs to adapt to varying illumination conditions, while also handling challenges such as similar vehicles, different occlusions, and image blur caused by camera shake or fast vehicle motion. Considering the road surveillance pattern and the vehicle’s appearance and motion changes, this paper proposes a Siamese vehicle tracker for road surveillance, namely SiamVTRS. It achieves a pragmatic and efficient combination of SiamCAR, a Kalman filter with camera motion compensation (CMC), and an adaptive template update mechanism. Additionally, SiamCAR is appropriately optimized and equipped with the average peak-to-correlation energy (APCE) assessment, which can perceive and suppress the common distractions in road scenarios. Furthermore, a re-capture strategy considering the target prior’s state is proposed to assist tracking when the target is missing or faces severe distractions. It performs global detection of candidate vehicles based on the target’s prior location and appearance state, aiming to rapidly re-locating the target. Extensive experiments are conducted on the UA-DETRC, DTB70, and UAV123 datasets, where the success rates of SiamVTRS on vehicle sequences reach 79.2%, 88.2%, and 91.8%, respectively. Compared with other mainstream trackers, SiamVTRS achieves competitive performance and demonstrates superior vehicle tracking in diverse road scenarios.