Adaptive Target Detection Algorithm Based On Combining Multimodal Data

Existing intelligent transport systems suffer from long response times in accident detection and processing, and are unable to provide timely warning and emergency response. To address this challenge, we propose a target behaviour recognition algorithm based on multimodal data understanding. The algorithm aims to improve the data processing and response speed of intelligent transport systems. The algorithm extracts the key information of the target from the multimodal data in the traffic scene. It also introduces an adaptive multi-task learning model combined with an attention mechanism. The mechanism achieves multi-faceted importance scoring of the input information by processing multiple query vectors in parallel. The algorithm automatically adjusts the model focus according to different scores for real-time changes in the traffic scene. The algorithm computes the dot product and weights between the query vectors and the input features, and returns a message for each query that incorporates different modal data and highlights features related to the target behaviour. Based on this information, the algorithm improves the speed and accuracy of target behaviour recognition processing. Experimental results show that the algorithm significantly improves the processing speed and accuracy of target behaviour recognition. It helps to improve the adaptability and responsiveness of the intelligent transport system to complex traffic scenes. For specific data, our algorithm reduces processing time by 0.65 s and improves recognition accuracy by 6% compared to traditional methods.