Multiple Object Detection and Tracking Using Deep Learning Framework with Non-Maximum Suppression

Multiple object detection and tracking involves identifying and locating numerous objects within a sequence of images or video frames and maintaining their identities across frames.This process is significant for applications like surveillance and autonomous vehicles.However accurately detecting and tracking multiple objects in dynamic environments is challenging because of occlusions, overlapping trajectories, and varying object sizes which results in tracking failure and misidentification.In this research, the Convolutional Neural Network with Non-Maximum Suppression (CNN-NMS) is proposed to detect and track multiple objects accurately using the VGG16 model.Using a CNN-NMS enhances object detection by leveraging the CNN's ability to extract detailed features whereas NMS refined the outcomes by removing redundant overlapping bounding boxes.In CNN-NMS, the VGG16 is applied to extract features with incorporating linear layer and sigmoid activation which predicts class labels and bounding box offsets.The proposed CNN-NMS achieves a high MOTA of 80.05%, 85.72%, 79.12%, 81.27%, and 82.47% using MOT17, open image bus truck, MOT15, 16, and 20 datasets compared to existing methods like You Only Look Once Tracker and RetinaMOT.