Pedestrian trajectory prediction is crucial in driverless applications. To accurately predict the high-quality trajectory of pedestrians, it is necessary to consider the reasonable social interaction and the spatiotemporal relationships between pedestrians. Previous methods couldn’t accurately capture the social features of pedestrians in realistic congested situations and extract spatiotemporal interaction features with high computation. Therefore, in this paper, a novel prediction model is proposed, called the Social Knowledge-guided Graph Attention Convolutional Network (SKGACN), which aims to address the social interactions and the spatiotemporal relationships between pedestrians with low computational requirements. Specifically, the social knowledge-guided graph attention mechanism fully considers multiple information relative to pedestrians to capture their social interaction. For spatiotemporal interactions, an improved Temporal Convolution Network (TCN) model is use as it can parallelize the processing times to get as higher efficiency compared to traditional models. Compared to the state-of-the-art methods, we evaluate our proposed method after applying it on two public datasets (ETH and UCY). The experimental results show that our method performs better in terms of Average Displacement Error (ADE) and Final Displacement Error (FDE) metrics.

Pedestrian trajectory prediction is crucial for enabling dynamic obstacle avoidance in social robots. Variational Autoencoders (VAE) have shown potential in predicting multi-modal distributions of future pedestrian trajectories. However, standards VAE struggle to generate accurate future trajectories, and existing prediction methods often overlook the relationships between pedestrian groups. This paper introduces a novel prediction model, called the Learning Autoencoder Diffusion Model (LADM) of Pedestrian Group Relationships for Multimodal Trajectory Prediction, which takes into account pedestrian group relationships, enhancing the accuracy of multimodal distribution trajectory prediction. In the LADM framework, each pedestrian is assigned to their most probable group through a learning process, and the interaction relationships between pedestrians and groups are determined using a pedestrian-group interaction module. To improve the quality of generated future trajectory distributions, we propose the Autoencoder Diffusion Model, the VAE functions as a generator and a diffusion model acts as a refiner. We evaluate our proposed method on two public datasets (ETH and UCY) and compare it to state-of-the-art methods. Experimental results demonstrate that our approach outperforms existing methods in terms of Average Displacement Error (ADE) and Final Displacement Error (FDE) metrics.

Predicting how pedestrians will move in the future is crucial for robot navigation, autonomous driving, and video surveillance. The complex interactions among pedestrians make it difficult to predict their future trajectory. Previous studies have primarily focused on modeling the interaction features of all pedestrians in the scene. However, this approach often results in an abundance of irrelevant interactions and overlooks the time-dependent characteristics. To address these issues, we propose a spatio-temporal adaptive graph pooling network (STAGP) for pedestrian trajectory prediction. STAGP adopts adaptive graph pooling to explicitly model interactions between pedestrians, redundant interactions and establishing directed interactions. In addition, we utilize spatio-temporal attention to extract temporal features of pedestrian interactions. For the prediction of future trajectories, we use a time-extrapolator convolutional neural network (TXP-CNN). ETH and UCY datasets were used to evaluate STAGP. Comparing STAGP to the values, the experimental results indicate that it is competitive in terms of ADE and FDE metrics.

Since Vision Transformers (ViTs) are introduced into computer vision, they have developed rapidly in a variety of visual tasks. Recently, they have been gradually applied to visual tracking. The Transformer can adaptively capture the global similarity comparisons of target objects and search regions, which has achieved competitive performance results. However, Transformer architectures often require a large amount of training data and computing resources, and lack prior knowledge of inductive biases that existed in images. The advantages of convolutional neural networks (CNNs) in extracting local similarities are not fully exploited. To resolve these problems, we propose a lightweight tracking architecture, combining CNN and Transformer in the feature fusion stage. Specifically, Local-Global Feature Interaction (LGFI) module and Feature Cross-Fusion (FCF) module are the key components in our approach. In the LGFI module, the proposed method includes a Transformer global information network and a Transformer-like CNN local information network for simultaneous global scope dependency establishment and local feature similarity enhancement, then aggregates their feature results together. In the FCF module, the proposed method includes a multi-head cross-attention and a convolutional feedforward network for feature fusion of templates and search regions. Finally, we use the classification and regression head to predict the exact location of the target. Extensive experiments demonstrate that, our method achieves better tracking performance than the baseline method, when both methods are trained with fewer data. Meanwhile, without any extra training data, the proposed method also obtains comparable results with other state-of-the-art trackers on six challenging benchmarks, including GOT-10k, LaSOT, TrackingNet, OTB100, UAV123, and NFS. Furthermore, our model is lightweight compared with the baseline method, with fewer parameters and lower FLOPs, while running at real-time speed.

Recently, there has been a surge of interest in using one-shot methods for multi-object tracking (MOT). These methods use a single network to produce both object detection results and embedding features simultaneously, achieving a balance of accuracy and speed. However, it is hard for the backbone network to extract high-quality feature information in complex scenes such as complex backgrounds or occlusions. In addition, most methods rely on identical rules to fuse appearance and motion information during the data association phase, which may fail when the target is briefly obscured or lost. In this work, we propose a novel multi-object tracker FSTrack that aims to address the challenges mentioned above. Our proposed solution incorporates feature enhancement and similarity estimation techniques to improve model performance. Specifically, we introduce the efficient channel attention module into the backbone network to facilitate better information interaction between channels and enhance representation capability. Furthermore, we propose a novel similarity matrix that combines appearance distance and DIoU distance of the target, resulting in superior association accuracy and fewer identity switching times. Experimental results on the MOT benchmarks, MOT17 and MOT20, demonstrate the superiority of our approach, especially the significantly improved tracking continuity.