In the human-robot collaborative manufacturing environment where humans and robots coexist, safety protection of human operators in real time is of paramount importance. This paper presents an approach for real-time active collision avoidance in augmented environment, where virtual 3D models of robots and real camera images of operators are used for monitoring and collision detection. A cost-effective depth camera is chosen for surveillance of any mobile foreign objects, including operators, which are not presented in the virtual 3D models. Two redundant Kinect sensors using structured light are used as the depth cameras for better area coverage and for eliminating possible blind spots in the surveillance area. Collision detection is performed by minumum distance. Processing applied on depth images includes background removal, filtering, labeling and points cloud generation. A prototype system is developed and linked to robot controllers for real-time robot control, with zero robot programming. According to the result of collision detection, it can alert an operator, stop a robot, or even move a robot away from an approaching operator. The results of a case study show that this approach can be applied to real-world applications such as human-robot collaborative assembly to safeguard human operators.