Visual Place Recognition for Water Environment Based on Semantic and Sequential Constrains

Recognizing places in water environment poses unique challenges due to the scarcity of distinctive features, significant variations caused by weather conditions, and the presence of moving objects. Existing visual place recognition (VPR) methods designed for land-based scenes often yield suboptimal results when applied to water environment. To address these difficulties, this paper proposes a visual place recognition method that leverages semantic and sequential constraints. The model incorporates a semantic constraint algorithm that effectively mitigates the issue of invalid and interfering features by directing focus towards key features. Additionally, a descriptor matching algorithm is introduced, which utilizes sequential constraints and hierarchical matching of descriptors to better capture and adapt to temporal information. Through comprehensive training and testing on real-world image sequences of marine scenes, the proposed method outperforms other state-of-the-art VPR algorithms. It achieves substantial improvements of 0.09, 0.22, 0.32, and 0.15 in Recall@1 compared to the second-best models, demonstrating its superior performance in water scene recognition.