Lower limb assistive devices have shown potential to restore mobility to millions of individuals with walking impairments; however, their success depends on whether they can be controlled safely, reliably, and intuitively with user-friendly sensors. To assist the user's walking patterns, many devices implement finite-state controllers which rely on accurate estimation of the current gait phase (e.g. Stance, swing) of one or both legs. Bilateral gait segmentation is especially important for restoring natural interlimb coordination, which contributes to device safety and efficiency. Most existing techniques for gait segmentation use ground contact, device-embedded, or body-worn sensors with threshold or machine learning-based algorithms. They have been effective at identifying the state of the ipsilateral (i.e. Sensor-side) leg but can become inconvenient for bilateral gait segmentation because they often require many sensors and are more sensitive to sensor placement. Therefore, we present a proof of concept for a novel approach to bilateral gait segmentation using a thigh-mounted inertial measurement unit (IMU) and depth sensor with the contralateral leg in its field of view. We extracted two features, ground and shank angle, from the depth data and developed a sensor fusion strategy to predict contralateral heel contact and ipsilateral toe off with accuracy approaching that of a setup with bilateral thigh and shank IMUs. By using computer vision to estimate the state of both legs, we introduce a new technique for bilateral gait segmentation which could make assistive devices more user-friendly, safe, and functional.