Human-Intention Prediction with Visual-Language Model

Human-intention prediction is an important part of human-machine interaction wildly utilized in industrial intelligent systems. In recent years, large language models have expanded to the image task with outstanding performance, leading to an increasing attraction to the application of multimodal Large Language Models. However, the exploration of visual-language models in human-intention prediction is still limited. To address this gap, this paper investigates the effectiveness of visual-language models in predicting human intentions and successfully transfers the knowledge in LLMs to downstream classification tasks. Finally, this paper takes traffic scenarios as an example to validate the feasibility of the video-LLaMA model in predicting pedestrian behavior intentions.