基于传感器非视觉信息的无人机目标朝向识别

Target Orientation Recognition Approach of UAV Based on Non-Visual Information from Sensors

很多传统方法仅利用视觉外观特征不足以实现理想的无人机面向人体评估结果.针对此问题,结合非视觉信息,提出一种对目标身体朝向进行实时估计的传感器信息融合方法.首先,对每种人体朝向训练一组视觉外观模型,每一种模型对应一系列无人机采集的遥测数据非视觉信息;其次,使用隐性动态条件随机域提取传感器非视觉信息间的交互信息,并利用Akaike信息准则对提取特征进行评估选择;然后,利用提取的高辨识度特征匹配每种视觉外观模型,得到目标朝向结果;最后,根据识别的目标结果自动调整无人机面向目标的正面.实验结果表明,与传统方法相比,所提方法的目标朝向识别准确率明显提高,达到90%左右.

Many traditional methods only use low-level visual appearance features,which are not enough to achieve ideal human body evaluation results for UAV.Aiming at the problem,a sensor information fusion method for real-time estimation of target body orientation is proposed based on non-visual information.Firstly,a set of visual appearance models are trained for each human orientation,and each model corresponds to a series of non-visual information of telemetry data collected by UAV.Then,hidden dynamic conditional random fields are used to extract non interactive visual information between the sensor information,and Akaike information criterion is adopted to evaluate the selection of feature extraction.And each visual appearance model is matched by the extracted high recognition feature,and the target orientation results are obtained.Finally,according to the target recognition results,the front of the UAV is automatically adjusted.The experimental results show that compared with the traditional method,the proposed method has significantly improved the accuracy of target orientation recognition,reaching about 90%.