Aiming to solve the aggregation problem of traditional uniform oriented FAST and rotated BRIEF（ORB） features in visual simultaneous localization and mapping（SLAM）， a uniform FAST corner extraction method is designed， which is based on grid division and laying to determinate key points. Furthermore， a corresponding loop detection method is designed based on uniform distribution of ORB features combined with brute force matching. Experiment results compared with BoW-based loop detection algorithms show that the proposed algorithm can significantly improve the accuracy of loop detection. Furthermore， a robot operating system（ROS） based loosely coupled semi-direct SLAM system is designed， which combine the uniform ORB feature loop detection module with direct sparse odometry（DSO）. The experimental results show that the proposed system has high map construction performance.