ABSTRACT: The issue of choosing appearance features for multiple object tracking (MOT) in urban scenes is addressed in this paper. Numerous features have been employed for MOT over the years. Whether some of them are superior to others is unclear, though. Colour histograms, histograms of oriented gradients, deep features from convolutional neural networks, and re-identification (ReID) features are examples of frequently used features. In this study, we evaluate the performance of these features in urban scene tracking scenarios to distinguish objects from a bounding box. Several affinity measures, including the Rank-1 counts, the cosine similarity, the L1, L2, and Bhattacharyya distances, are also evaluated for their effect on the discriminative power of the features. . Results from several datasets demonstrate that, regardless of the detector quality, features from ReID networks are the best at differentiating between instances. Colour histograms may be chosen in the absence of a ReID model if the detector has a good recall and few occlusions; otherwise, deep features are more resistant to detectors with lower recall. An picture's colour histogram shows how the colours are distributed throughout the image. The number of pixels in each type of colour and the various colour variations are displayed. Re-identification is a general term for any process that re-establishes the relationship between data and the subject to which the data refer