摘要
为了较好地对视频中交通事件进行语义理解,减少底层特征与高层事件间的语义鸿沟,提出一种具有通用性的交通事件识别框架.首先,把事件分为单车行为、人车交互行为与车车交互行为;然后,根据语义层次,按基本语义单元、基础语义事件和高级语义事件的顺序对交通事件进行识别,并根据逻辑化的自然约束语言(NCL)的规则,给出了交通事件的语义表达形式,通过对高级语义事件建立隐马尔可夫模型(HMM)得到事件的高级语义,在一定程度上跨越了语义鸿沟;最后,通过实验验证了提出的方法.
In order to acquire semantic comprehension of traffic events more clearly, and to reduce the semantic gap between low-level feature and high-level events, a universal recognition framework for traffic events was developed, which firstly classified events into three types： single vehicle behavior, human-vehicle interaction behavior and vehicle-vehicle interaction behavior. Then the traffic events were identified in order of basic semantic unit, basic semantic event, and high-level semantic event according to semantic level. Moreover, a semantic expression in traffic event was proposed based on the rule of logical natural constraint language （ NCL）, and the high-level semanteme of an event was got by establishing hidden Markov model （HMM） to a high-level semantic event, thus diminishing the semantic gap to a certain extent. At last, the method proposed was proved by experiment.
出处
《应用科技》
CAS
2013年第2期5-10,共6页
Applied Science and Technology
基金
黑龙江省交通厅重点项目(2012JT016)
关键词
交通事件
识别框架
自然约束语言
语义理解
语义表达
隐马尔可夫
traffic event
recognition framework
natural constraint language （NCL）
semantic comprehension
semantic expression
hidden Markov model （HMM）
