Multiple object tracking using deep learning and machine learning techniques
Mahesh Ratnaparkhe, U. Sivaji, Sachin Turreraa, Sirra Yashwanth · 2024
This chapter presents a robust framework for multiple object tracking using a combination of deep learning and machine learning techniques. The methodology comprises four phases: preprocessing and model initialization, object detection and feature extraction, track management and association, and tracking visualization and output. Key components include the integration of YOLOv7 for real-time object detection, conversion of pretrained weights, and the incorporation of DeepSORT for track association using Kalman filters and appearance similarity metrics. TensorFlow serves as the underlying framework for deep learning operations. The chapter seamlessly handles datasets for both object detection and tracking, leveraging common benchmarks such as COCO, VOC, and MOT datasets. The results showcase the effectiveness of YOLOv7 and DeepSORT in accurately identifying and tracking objects, even in scenarios with occlusion. The dynamic configuration of tracked classes adds flexibility to the system, allowing users to tailor object tracking based on specific requirements. The fusion of cutting-edge technologies enables the project to excel in various applications, from video surveillance to behavior analysis.