Object Detection and Tracking Using DeepSORT

Abstract

The rapid development of algorithms for image detection has resulted in broad safety applications, for example, face recognition and monitoring. However, monitoring in real-time is quite difficult, particularly in busy areas, where the individual may be partially or fully concealed for a period of time. Therefore, this chapter aims to build the object tracking system with the DeepSORT architecture for crowd surveillance. This system not only detects a human being in real time but also uses the information it has learnt, in contrast to the object detection frameworks like CNN, to monitor a human's path until they leave the frame. Using the DeepSORT framework, the trajectory of an item may either be detected in real time or even in a background based on an existing record. This framework will be trained using a huge dataset to track people's motion depending on the individual's speed, distance and fitness. The DeepSORT algorithm, the top algorithm in object identification and tracking, is very powerful and quick. Many issues have been identified with the security mechanism for object tracking. Occlusion is one of the main problems in the pursuit of objects. This problem occurs in a busy location with another object moving extremely quickly into and out of the picture (i.e., a bustling Zebra crossing). If an item is present inside a frame, the trajectory of an individual may be incorrectly identified.