Visual odometry for outdoor environment using a downward-tilting camera and self-shadow removal algorithm

Stereo vision based visual odometry faces few challenges when it is performed in the outdoor environment. The first problem is the inaccurate depth values which result in poor distance estimation. This problem can be solved by tilting a stereo camera downward and using features from the ground texture. Second, self-shadow made by the sun can causes errors in estimating the robot movement since it looks stationary in the robot's perspective. This was resolved by estimating the sun ray direction and excluding the self-shadow area in the camera image. This process can be conducted by using only a first image that the robot grabs. These two algorithms were proven to be successful in improving the accuracy of visual odometry in the outdoor environment.