顾及短基线同轴约束的指向性观测相机原型

针对相机大视场感兴趣目标分辨率不足的问题，提出短基线同轴约束模型并设计主从相机原型,大视场相机用于监视整个视场，主动相机用于对目标区域进行指向性高清观测。基本过程为：（1）利用短基线同轴约束模型简化相机外参矩阵,并构建大视场相机与主动相机间的映射关系，通过三角函数计算主动相机初始控制参数；（2）在近距离场景下，对主动相机控制参数进行补偿。实验结果表明，相机只需一次离线标定即可适应各种场景。观测目标与大视场相机距离3～10 m的近距离场景下，目标在主动相机图像中的实际位置与理论位置误差在30像素以内；远距离场景的有效距离内，误差在6像素左右。计算单一目标点对应主动相机控制参数的时间不超过0.2 ms。原型对目标场景、目标深度无依赖性，且用于较远目标观测时相对于其他方法在精度与时效性方面具有较高优势。

Prototype of Directional Observation Camera Considering Coaxial Constraint of Short Baselines

  Objectives  Considering the insufficient resolution of interested targets in a large field of view camera, a coaxial constraint model of short baselines is proposed, and a master-slave camera prototype is designed. A large field of view camera is used to monitor the entire field of view, and the directional high-definition observation is carried out by an active camera.  Methods  The yaw angle and pitch angle of the active camera are solved by using the image point of the target center point in the large field of view camera image. Firstly, the mapping relationship between the large field of view camera and the active camera is constructed, and the external parameter matrix of the camera is simplified by using the coaxial constraint model of short baselines. The initial control parameters of the active camera are solved through trigonom?et?ric functions. Secondly, the pitch-angle compensation values of close-range scenes are calculated according to the installation position of the prototype.  Results  Experimental results show that(1) The compensation for the control parameters of the active camera in close-range scenes can improve the control accuracy to a certain extent, and the compensation effect in the central area is better. (2) In close-range scenes where the distance between the observation target and the large field of view camera is 3?-?10 m, the prototype can actively observe the details of the target. The actual and ideal position error of the target in the active camera image is less than 30 pixels, and the horizontal direction error can be ignored. (3) In the outdoor long-distance scenes, the prototype can actively and accurately observe the target details within the effective distance. As the depth of the observed target increases, the error decreases gradually. In the range of 40?-?50 m, the system error caused by steering engine control accuracy and baseline length becomes the main error, and the overall error is maintained at about 6 pixels. (4) The time for prototype to calculate the control parameters of the active camera of a single target point is within 0.2 ms, which can meet the real-time requirement. (5) The algorithm solves the control parameters of the active camera without scene dependence. It can control the rotation of the active cam?era with high precision and strong adaptability in different scenes. (6) When the large field of view cam?era is used to observe the far target at low resolution, the target cannot be located accurately in the im?age. Improving the resolution of the large field camera is helpful to improve the accuracy of active camera control parameters and effective observation distance.  Conclusions  The directional high-definition observation of interested regions in the large field of view can be realized through the prototype. The camera part is modularized, and only one offline calibration is needed to adapt to different scenes and target depth. The accurate and fast target observation is realized. Compared with oth?er methods, it has higher accuracy and timeliness with better applicability. Additionally, the accuracy and the observation distance can be increased by properly improving the resolution of the large field of view cam?era.