A number of advantages of unicondylar arthroplasty (UKA) over total knee arthroplasty in patients presenting osteoarthritis in only a single compartment have been identified in the literature. However, accurate implant positioning and alignment targets, which have been shown to significantly affect outcomes, are routinely missed by conventional techniques. Computer Assisted Orthopaedic Surgery (CAOS) has demonstrated its ability to improve implant accuracy, reducing outliers. Despite this, existing commercial systems have seen extremely limited adoption. Survey indicates the bulk, cost, and complexity of existing systems as inhibitive characteristics. We present a concept system based upon small scale head mounted tracking and augmented reality guidance intended to mitigate these factors. A visible-spectrum stereoscopic system, able to track multiple fiducial markers to 6DoF via photogrammetry and perform semi-active speed constrained resection, was combined with a head mounted display, to provide a video-see-through augmented reality system. The accuracy of this system was investigated by probing 180 points upon a 110×110×50 mm known geometry and performing controlled resection upon a 60×60×15 mm bone phantom guided by an overlaid augmented resection guide that updated in real-time. The system produced an RMS probing accuracy and precision of 0.55±0.04 and 0.10±0.01 mm, respectively. Controlled resection resulted in an absolute resection error of 0.34±0.04 mm with a general trend of over-resection of 0.10±0.07 mm. The system was able to achieve the sub-millimetre accuracy considered necessary to successfully position unicondylar knee implants. Several refinements of the system, such as pose filtering, are expected to increase the functional volume over which this accuracy is obtained. The presented system improves upon several objections to existing commercial