The paper describes a computer vision method for estimating the clinical gait metrics of walking patients in unconstrained environments. The method employs background subtraction to produce a silhouette of the subject and a randomized decision forest to detect their feet. Given the feet detections, the stride and step length, cadence, and walking speed are estimated. Validation of the system is presented through an error analysis on manually annotated videos of subjects walking in different outdoor settings. This method is significant as it provides clinical therapists and non-specialists the opportunity to record from any camera and obtain high accuracy estimates of the clinical gait metrics for subjects walking at outdoor or at-home locations.