Abstract: The main research focus in the Computer Vision Group at JPL is visual navigation for unmanned vehicles, including ground, sea, and air vehicles on Earth and rovers and landers in planetary exploration. This talk will start with a brief overview of the range of our activities, then present recent work in two areas. First, the main thrust of our NASA-funded work currently is on vision systems for safe and precise landing; I will give an overview of work on these problems for eventual missions to the moon, Mars, comets, asteroids, and Titan. Second, we are working on detecting people around ground vehicles for safe navigation, using classification and tracking applied to 3D point clouds from dense stereo vision. Our system performs dense stereo, visual odometry, and pedestrian detection at 5 Hz on 1024×768 imagery with one quad core embedded Pentium, with ROCs that compare favorably to other recent work that runs very slowly off-line.