SORT with Depth Image Based Pedestrian Tracking Robots

Aiming at the problem of frequent position changes, crossing, and disappearance of pedestrians in the image coordinate caused by the low camera angle of view on the tracking robot, we have made a series of adaptive improvements to the SORT (Simple Online and Realtime Tracking), an online multiple object tracking algorithm, furthermore propose the SORT-DI (Simple Online and Realtimes Tracking with Depth Image). We use the RGB-D camera to obtain the position of the target in the robot coordinate system and use a simple and fast filter to calculate the difference between image features such as gradient features and color features, which solves the problem that the SORT is easy to identify swiches. This solution can also run faster and with low latency on the CPU and achieve the online tracking of designated pedestrians by the robot in large-scale pedestrians tracking scenarios.