Predicting the locomotion intent of humans is important for controlling assistive robots predictively. Previous studies have researched assistive walking on structured terrains, but few studies consider rough terrains. Human intent on rough terrains is more difficult to predict because the transition happens every step. To predict foot placements of humans on rough terrains, this paper fuses sequential 3D gazes and environmental context. The 3D gaze is assumed to be the intersection point of the line of sight measured by an eye-tracker and the environmental point cloud measured by an RGBD camera. The sequential 3D gazes and environmental context are fused based on an RGBD SLAM algorithm. Then the closest segmented terrain to the center of sequential 3D gazes is regarded as the most possible foothold area at the next step. Six able-bodied subjects participate in the experiments, and their foot placements can be accurately predicted 0.5 step ahead. With environmental context, the average classification accuracy and the distance error of predicting the foot placements can achieve 97% and 0.11 m. The user-dependent time window can further decrease the prediction error to 0.086 m. Without environmental context, the distance between the gaze center and the actual foot placement is higher than 0.18 m. Hence, both gazes and environmental context are important to predict human intent on rough terrains.