Camera calibration methods include: traditional camera calibration method, active vision camera calibration method, camera self calibration method, zero distortion camera calibration method.
The traditional camera calibration method needs to use the calibration object with known size. By establishing the correspondence between the points with known coordinates on the calibration object and their image points, the internal and external parameters of the camera model can be obtained using a certain algorithm. According to the different calibration objects, they can be divided into three-dimensional calibration objects and planar calibration objects. The three-dimensional calibration object can be calibrated by a single image with high calibration accuracy, but it is difficult to process and maintain the high-precision three-dimensional calibration object. The planar calibration object is simpler to make than the three-dimensional calibration object, and its accuracy is easy to ensure, but two or more images must be used for calibration. The traditional camera calibration method always needs a calibration object in the calibration process, and the manufacturing accuracy of the calibration object will affect the calibration results. At the same time, some occasions are not suitable for placing calibration objects, which also limits the application of traditional camera calibration methods.
At present, self calibration algorithms mainly use camera motion constraints. The motion constraint of the camera is too strong, so it is not practical in practice. Using scene constraints is mainly to use some parallel or orthogonal information in the scene. The intersection of spatial parallel lines on the camera image plane is called vanishing point, which is a very important feature in projective geometry. Therefore, many scholars have studied camera self calibration methods based on vanishing point. The self calibration method is flexible and can calibrate the camera online. But because it is based on absolute conic or surface method, its algorithm robustness is poor.
The camera calibration method based on active vision is to calibrate the camera with some motion information of the camera known. This method does not need the calibration object, but needs to control the camera to do some special movements. The camera's internal parameters can be calculated using the particularity of this movement. The advantages of the camera calibration method based on active vision are that the algorithm is simple, and the linear solution can often be obtained, so the robustness is high. The disadvantages are that the cost of the system is high, the experimental equipment is expensive, and the requirements for experimental conditions are high, and it is not suitable for situations where the motion parameters are unknown or uncontrollable.
Zero distortion camera calibration method is to establish the mapping relationship between LCD pixels and camera sensor pixels, and determine the view point position of each camera pixel on the LCD, with LCD screen as the reference and phase-shift grating as the medium. The lens makes the field of view of the camera on the LCD non rectangular. In this distorted field of view, an embedded virtual sensor can be constructed, and the same number of pixels can be maintained. In this way, each virtual pixel must fall within an arbitrary quadrilateral formed by four adjacent viewpoints. The brightness of virtual pixels will be determined by weighted interpolation of the brightness of these four points, and is independent of other pixels. The original image is resampled (four times multiplication and four times addition) by using the set of four weighting coefficients, and the zero distortion output image can be obtained. The RGB three channels of the color camera are processed separately, but if a common virtual sensor is selected, the resultant color image will be zero distortion and zero color difference. The position error of each pixel (physical view point and virtual point) is zero mean, and the mean square error can be less than 1/1000 pixel pitch.
