上传了初稿中所用的所有图片

sot/undergraduate/content/abstracten.tex

@@ -1,41 +1,7 @@
 %!TEX root = ../csuthesis_main.tex
-\keywordsen{Digital Twin\ \ Autonomous Driving\ \ Reinforcement Learning\ \ Simulation System; \ \ Decision Optimization}
+\keywordsen{Autonomous Driving;\ \ Visual Perception;\ \ Target Tracking;\ \ Intention Recognition; \ \ Carla Simulation}
 \begin{abstracten}
 
-    With the rapid development of artificial intelligence and autonomous driving technology, traditional 
-    testing and validation methods can no longer meet the demands of complex and variable driving 
-    environments. Digital twin technology, as an emerging simulation method, provides an efficient and safe 
-    testing platform for autonomous driving systems by constructing virtual models that correspond to the 
-    real world. Digital twins can reflect the state of physical entities in real-time, supporting precise 
-    simulations of various aspects such as vehicle dynamics and environmental changes, thereby providing 
-    rich data support for the training and optimization of autonomous driving algorithms. Reinforcement 
-    learning, as a self-learning intelligent algorithm, can optimize decision-making processes in constantly 
-    changing environments. Combining digital twins with reinforcement learning can not only accelerate the 
-    development and validation of autonomous driving systems but also enhance their adaptability and safety 
-    in complex scenarios. Researching a digital twin-based reinforcement learning simulation system for 
-    autonomous driving has significant theoretical significance and practical application value.
-
-    This article aims to explore the design and implementation of a digital twin-based reinforcement 
-    learning simulation system for autonomous driving. With the rapid development of autonomous driving 
-    technology, traditional testing and validation methods can no longer meet the increasingly complex 
-    driving environments and diverse driving scenarios. Digital twin technology, as an emerging simulation 
-    method, provides an efficient and safe testing platform for autonomous driving systems by constructing 
-    virtual models corresponding to the real world. This article first reviews the development history of 
-    autonomous driving technology, analyzes the basic concepts of digital twins and their application 
-    potential in the field of autonomous driving. It delves into the basic principles of reinforcement learning 
-    and its importance in autonomous driving, emphasizing the necessity of optimizing autonomous driving 
-    decisions through reinforcement learning algorithms.
-
-    On this basis, the article proposes a simulation system framework that combines digital twins and 
-    reinforcement learning, detailing the system's architectural design, functional modules, and 
-    implementation process. By constructing a digital twin model of a real environment, the system can 
-    simulate a large number of driving scenarios in a virtual environment, thereby accelerating the 
-    reinforcement learning training process. Experimental results show that the system has significant 
-    advantages in improving the accuracy and stability of autonomous driving decisions. The article also 
-    discusses the challenges and future development directions of the system in practical applications, 
-    pointing out solutions to issues such as data scarcity and convergence, providing references for 
-    subsequent research. The digital twin-based reinforcement learning simulation system for autonomous 
-    driving not only offers new ideas for the validation and optimization of autonomous driving technology 
-    but also provides valuable practical experience for researchers in related fields.
+    With the rapid advancement of autonomous driving technology, enhancing perception and decision-making capabilities in complex traffic environments has become a crucial research focus. This study designs a vision-based target tracking and intention analysis system based on the Carla simulation platform to address the challenges of balancing accuracy and real-time performance faced by traditional tracking and behavior prediction algorithms in dense traffic scenarios. The system integrates the DeepSORT multi-object tracking algorithm, combining Kalman filtering and deep appearance feature extraction to achieve accurate and continuous tracking of dynamic targets. Additionally, a lightweight intention recognition mechanism based on physical modeling is developed, which analyzes the speed and relative position of targets to infer behavior trends such as "approaching," "departing," and "dangerous approach" in real-time. Experimental results show that the system maintains strong robustness and real-time performance, especially under conditions of occlusion and rapid motion in complex urban traffic environments. The proposed approach effectively improves the perception intelligence of autonomous vehicles, providing reliable support for decision-making and path planning. This research offers new insights and technical solutions for vision-based target tracking and intention analysis, with promising prospects for practical applications and further development in the field of intelligent transportation systems.
 
-\end{abstracten}
+\end{abstracten}