Monocular Visual Odometry - How to get the relative coordinate of an image according to the previous frame

Hi,

I'm here to ask a question about recovering the position of a robot from a frame to another, using images taken by a camera.

My process is simple, i have image taken at each frame.

I recover the features (corners + blobs), match them from a frame to another ...

At the end, i estimated my essential matrix with these points, and finally my rotation matrix and translation vector.

Then it seems to be not work, but i don't really know why.

                [E, epipolar] = estimateEssentialMatrix(location_l_p, location_l_a, stereo_params.CameraParameters2);
                % Find epipolar inliers -> Matches which follows the
                % epipolar constraint, avoiding using matches from the
                % borders. Normally, i don't need it.
                location_l_p = location_l_p(epipolar, :);
                location_l_a = location_l_a(epipolar, :);
                
                % 1 
                % Get the relative position of the camera in the actual
                % image using the [R] rotation matrix, [t] translation
                % vector and the Essential matrix of the previous left
                % image.
                
                [orientation, location] = relativeCameraPose(E, stereo_params.CameraParameters2, location_l_p, location_l_a);
                
                % Compute extrinsics of the second camera to get the parameters.
                [actual_R, actual_t] = cameraPoseToExtrinsics(orientation, location);
                
                
                R = R * actual_R;
                t = t + actual_t * R;
                
                

The things is that i want to return, at each frame, the position [x, y] in my coordinate system, using the previous coordinates and the actual R (Rotation Matrix) and T (translation vector).

Can you help me ?

PS : i verify my features selected, i used a lot of process and there good (from 100 features to 20 features : circle validation, selection uniform, epipolar ....)