Adaptively self-driving tracking algorithm based on particle filter

The promotion of autonomous vehicles is a decisive step to implement smart urban planning. The machine vision technique applied in the self-driving car can facilitate the car detecting and tracking other vehicles, pedestrians, lanes and traffic signs on the road, etc. This paper proposed an algorithm to track the vehicle with the adaptively changed scale. First, we use the tracker to obtain the vehicle candidates at each frame based on kernelized correlation filter. Next, an array of particles was created to represent different scales. Further, a new image feature representation based on integrated-color-histogram was proposed to insert the updated scheme concerning the particle filter algorithm. Last, we used one smooth method to make the scales change have its own memory to prevent it from violent variation. In the experiment section, we have chosen some pervasive tracker to analyze. The results showed that in the aspects of both accuracy and robustness, our proposed algorithm worked more properly compared with the other algorithm, by virtue of its minimal error relative to the data benchmark.