Unmanned vehicle systems, specifically Unmanned Air Vehicles (UAVs) and Unmanned Ground Vehicles (UGVs) have found potential use in both military and civilian applications. For many applications, unmanned vehicle systems are required to navigate in urban environments where obstacles with various types and sizes exist. The main contribution of this research is to offer vision-based path planning, collision avoidance, and target tracking strategies for Unmanned Air and Ground vehicles operating in urban environments. Two vision-based local-level frame mapping and planning techniques are first developed for Miniature Air Vehicles (MAVs). The techniques build maps and plan paths in the local-level frame of MAVs directly using the camera measurements without transforming to the inertial frame. Using a depth map of an environment obtained by computer vision methods, the first technique employs an extended Kalman Filter (EKF) to estimate the range, azimuth to, and height of obstacles, and constructs local spherical maps around MAVs. Based on the maps, the Rapidly-Exploring Random Tree (RRT) algorithm is used to plan collision-free Dubins paths. The second technique constructs local multi-resolution maps using an occupancy grid, which give higher resolution to the areas that are close to MAVs and give lower resolution to the areas that are far away. The maps are built using a log-polar representation. The two planning techniques are demonstrated in simulation and flight tests. Based on the observation that a camera does not provide accurate time-to-collision (TTC) measurements, two and three dimensional observability-based planning algorithms are explored. The techniques estimate both TTC and bearing using bearing-only measurements. A nonlinear observability analysis of state estimation process is conducted to obtain the conditions for complete observability of the system. Using the conditions, the observability-based planning algorithms are designed to minimize the estimation uncertainties while simultaneously avoiding collisions. The two dimensional planning algorithm parameterizes an obstacle using TTC and azimuth, and constructs local polar maps. The three dimensional planning algorithm parameterizes an obstacle using inverse TTC, azimuth, and elevation, and constructs local spherical maps. The algorithms are demonstrated in simulation. Lastly, a probabilistic path planning algorithm is developed for tracking a moving target in urban environments using UAVs and UGVs. The algorithm takes into account occlusions due to obstacles. It models the target using a dynamic occupancy grid and updates the target location using a Bayesian filter. Based on the target's current and probable future locations, a decentralized path planning algorithm is designed to generate suboptimal paths that maximize the sum of the joint probability of detection for all vehicles over a finite look-ahead horizon. Results demonstrate the planning algorithm is successful in solving the moving target tracking problem in urban environments.