It is an effective means to identify and track the target using the airborne electro-optical platform with high-resolution imaging equipment. However, in the actual combat process, it is challenging to meet the optimal observation conditions of nadir looking. Usually, it is necessary to quickly locate the remote target at a certain height, resulting in a large inclination of the photograph and a small angle between two observations. Furthermore, a motion camera's relative pose estimation is the key visual location technology based on an airborne electro-optical platform. Typical pose estimation methods include the vision-based pose estimation method and the vision-based pose estimation method with an inertial measurement unit (IMU). The latter introduces additional angle constraints based on the former. Under typical limited observation conditions such as a large inclination angle and small intersection angle, the camera's pose estimation accuracy is easily affected by the attitude angle error and the image point extraction error, which makes the target location accuracy difficult to meet application requirements. Therefore, it is of great significance to study target tracking and location under limited observation conditions to improve operational efficiency.

The unavoidable problem is that the measurement angle of the IMU drifts with time, and the calibration of the installation relationship between the IMU and the camera is cumbersome. Therefore, this paper proposes a ground target location method under the fixed installation of the IMU and the camera. This method does not require the installation relationship information between the calibration camera and the IMU. Firstly, two adjacent frames of images are extracted and matched through the scale-invariant feature transformation (SIFT) matching algorithm. Secondly, the IMU installed in the fixed link is used to provide the relative rotation angle information for the motion camera. Combined with the robust estimation algorithm random sample consensus (RANSAC) algorithm, the relative rotation and translation of two adjacent frames are estimated according to a relative rotation angle and four corresponding image points. Then, combined with the external parameters of the first frame image, the external parameter matrix of any frame can be obtained. Finally, the projection matrix is used to intersect the spatial location of the target.

Under the RANSAC framework, the smaller the minimum number of points required to solve the model, the same confidence level can be achieved with fewer iterations. Therefore, compared with other traditional algorithms, the algorithm proposed in this paper can significantly improve the computational efficiency under the same ratio of outliers (Fig. 3). In the simulation and flight experiment, the airborne electro-optical platform is 5 km away from the target and observes the target at an angle of 66.42°. Monte Carlo simulation analysis shows that the location accuracy of the proposed method is affected by the accuracy of pixel extraction (Fig. 5), relative rotation angle (Fig. 6), and the platform location (Fig. 7). Under the same pixel extraction error, the method in this paper shows apparently excellent performance. And the larger the intersection angle, the smaller the location error. According to the flight experiment, this method introduces the relative rotation angle constraint, which can more accurately estimate the relative rotation and translation of the camera than the traditional method and improve the location accuracy under limited observation conditions (Table 3).

Under limited observation conditions such as a long distance, a large inclination, and a small intersection angle, the accuracy of ground target location is challenging to meet the task requirements. Therefore, this paper proposes a ground target location method with the fixed installation of an IMU and a camera. When the IMU is fixedly installed with the camera, it can directly provide the relative rotation angle for the camera. What's more, it reduces the degrees of freedom of the relative pose estimation problem and the number of solutions. Under the RANSAC framework, the proposed method is more likely to find the appropriate solution from the same number of iterations. This method can significantly improve the calculation efficiency and the target location accuracy under limited observation conditions.