25 August 2020 in IEEE Access
In modern railway systems, video surveillance and machine vision analysis have been widely used to detect perimeter intrusions. For pan–tilt–zoom (PTZ) cameras, the machine vision system needs to detect adjustments in PTZ cameras and then automatically determine the new alarm region in real time. In this paper, we propose a deep multi-task learning based algorithm for simultaneous vanishing point (VP) detection and rail segmentation, which can identify camera adjustment from changes in VP, and then automatically determine the alarm region from segmented rails. The multi-task based neural network consists of a feature extraction base network and three sub-task networks. The first sub-task network is a convolution regression network for VP detection. The second sub-task network utilizes an encoder-decoder structure for vanishing region (VR, a fixed region centered on VP) segmentation. The third sub-task network shares the encoder-decoder structure with the VR segmentation task and is used for rail segmentation. The VR segmentation task is activated only at the training stage, serving as an auxiliary task to enhance feature learning ability and increase VP detection accuracy. To further improve the accuracies of VP detection and rail segmentation, low-level features is modulated by high-level semantic information before feeding to the decoder stage. With the help of shared feature extraction and auxiliary training, the proposed VP prediction method needs very small training dataset and outperforms other methods in both efficiency and accuracy.
4 December 2025 in IEEE Transactions on Industrial Informatics
Transformer-based multiobject tracking (MOT) methods have attracted significant attention from researchers. However, these Transformer-based models often suffer from suboptimal inference speeds due to their architectural complexities or other inherent issues, rendering them difficult to deploy in practical industrial applications. To address this challenge, we revisited the classic joint detection and tracking (JDT) paradigm and analyzed existing models. Drawing inspiration from Detection Transformer’s (DETR) object queries, which naturally encode object appearance features, we constructed a fast and novel JDT-type MOT framework named FastTrackTr by implementing an efficient interframe information transfer mechanism. This framework integrates three key technological innovations: a cross-decoder mechanism that implicitly incorporates historical trajectory information without requiring additional queries or decoders, a historical encoder and decoder pair for refining and utilizing historical feature representations, and a deterministic fixed-shape architecture that enables seamless TensorRT acceleration. Benefiting from these designs, our approach not only reduces the number of queries required for tracking but also avoids introducing excessive network structures, ensuring model simplicity while maintaining high accuracy. Experimental results show that our method achieves real-time tracking while maintaining state-of-the-art accuracy. On an NVIDIA RTX 4090 with an image size of 1333 × 800, it reaches 62.4 higher order tracking accuracy (HOTA) at 86.6 frames per second (FPS) on the DanceTrack dataset, outperforming other advanced Transformer-based methods. Furthermore, it excels on edge devices, such as the NVIDIA Jetson AGX Orin, where its lightweight variant achieves up to 59.2 FPS on 640 × 640 images, meeting real-time requirements for practical applications.