*Result*: Temporal aggregation for real-time RGBT tracking via fast decision-level fusion.

*RGBT tracking, which involves visual object tracking using both RGB and thermal infrared spectra, is a rapidly growing research field nowadays. However, the introduction of multiple modalities poses a challenge for real-time requirements. To address this issue, our method incorporates a lightweight decision-level fusion module (DFM). The DFM takes tracking results from individual modalities as input, residing in a low-dimensional manifold, which enables the network to be lightweight and efficient. Accordingly, the fusion of multi-modal information is achieved in a neglectable time, 0.0009s. In addition, recognising that the effectiveness of the fusion block depends on the quality of the input data, a temporal information aggregation module (TIAM) is introduced to integrate temporal cues into spatial information, producing more discriminative feature embeddings. Unlike the features extracted from the spatial-only models, TIAM reduces the impact of stationary distractions by highlighting moving content. Extensive experiments conducted on several challenging benchmarks, including VOT-RGBT2019, GTOT, RGBT210, LasHeR, and VTUAV, demonstrate the effectiveness of our method. Source codes are available at https://github.com/Zhangyong-Tang/TAAT. • A lightweight fusion module is implemented at the decision level. • A temporal aggregation module is proposed to combine the spatial and temporal clues. • Experiments demonstrate the efficiency and effectiveness of the proposed components. [ABSTRACT FROM AUTHOR]*