Tracking errors due to particles moving in and out of the focal plane are a fundamental problem of multiple particle tracking microrheology. Here, we present a new approach to treat these errors so that a statistically significant number of particle trajectories with reasonable length are received, which is important for an unbiased analysis of multiple particle tracking data from inhomogeneous fluids. Starting from Crocker and Grier's tracking algorithm, we identify particle displacements between subsequent images as artificial jumps; if this displacement deviates more than four standard deviations from the mean value, trajectories are terminated at such positions. In a further processing step, trajectories separated by a time gap Delta tau(max) are merged based on an adaptive search radius criterion accounting for individual particle mobility. For a series of Newtonian fluids covering the viscosity range 6-1300 mPa s, this approach yields the correct viscosity but also results in a viscosity-independent number of trajectories equal to the average number of particles in an image with a minimum length covering at least two orders of ... mehr magnitude in time. This allows for an unbiased characterization of heterogeneous fluids. For a Carbopol ETD 2050 solution we recover the expected broad variation of particle mobility. Consistent with the widely accepted structural model of highly swollen microgel particles suspended in a polymer solution, we find about 2/3 of the tracers are elastically trapped.