The use of robots in service and manufacturing industries has increased significantly in past decades. These robots share the same workspace with humans, so physical interactions between humans and robots occur. Thus, collision between a human and a robot is inevitable during these interactions, and safety measures must be provided to protect humans. In this study, a disturbance observer based on generalized momentum is proposed to efficiently detect a collision. The proposed algorithm of the disturbance observer allows any multi-DOF robot to effectively detect a collision without the use of additional sensors or mechanisms. Once a collision is detected, the robot can perform the appropriate reaction to minimize the impact on the human and the robot. Several reaction strategies are proposed for different situations. The proposed algorithm was implemented on a 7 DOF robot arm to verify its performance. Furthermore, the proposed reaction strategies were also demonstrated using the same robot arm.