This paper discusses the development, implementation and deployment of a computationally light, yet robust real-time collision avoidance system on a hover capable coaxial rotary wing Micro Air Vehicle (MAV). The real-time capability of the algorithm is demonstrated by performing all computations required for processing the depth camera image onboard at 13 Hz using a single core of a 1.6 GHz quad core processor based single-board computer. The primary sensor used is Microsoft Kinect which has an operating range of 0.5 – 3.5 m. The algorithm uses template matching feature of an open source library OpenCV to find a window through which a vehicle of specified dimensions and known speed can pass safely. The actuator control and active yaw stabilization is done using Navstik which is a Micro Navigation & Control hardware. Two test vehicles are built and equipped with this framework for proof of concept.