Method based on Separation Confidence Computation and Scale Synthesis Optimization for Real-Time Target Detection in Streetscape Videos

摘要

This study proposes a method for the real-time detection and recognition of targets in streetscape videos. The proposed method is based on separation confidence computation and scale synthesis optimization. First, on the basis of generalization in transfer learning, we combine a fine-tuning method suitable for non-convex optimization and adaptive moment estimation in high-dimensional space. Then, we dynamically adjust the learning rates of parameters on the basis of first and second gradient moment estimations. We establish the framework and implementation steps of the proposed method by organically combining regular term super-parameter generalization and hard-example mining technology. We use the proposed method to detect and recognize targets in streetscape videos with high frame rates and high definition. Furthermore, we experimentally demonstrate that the accuracy and robustness of our proposed method are superior to those of conventional methods.