Real-Time Multiple Object Detection and Tracking Using Adaptive Frame Cancellation

In image processing and computer vision, multi-object detection tracking (MODT) is an essential and widely studied area of inquiry.Predicting full tracklets for several objects in a video clip is the objective of the MODT problem.Occlusion and other comparable items are only two of the many obstacles that might hinder a tracking algorithm's Performance.However, in complex settings, the tracking performance of the present MODT algorithms based on the tracking-by-detection paradigm degrades due to increased ID shifts, tracking drifts, accuracy, and increased the computation time.The proposed algorithm is the real-time multiple object detection and tracking using adaptive frame cancellation.The suggested algorithm for enhancing MODT utilizing the Frames Cancellation Technique (FCT) demonstrates a substantial reduction in execution time compared to that of other MODT algorithms.To maintain the accuracy of tracking objects within an acceptable range while the frame cancellation rate escalates, it is necessary to implement negative feedback to enhance system tracking stability and thus improve accuracy.This feedback will enable the proposed algorithm to adjust the frame rate cancellation adaptively while maintaining a relatively constant overall accuracy, contingent upon the errors generated during execution.The Kalman filter residual is the most effective signal feedback for finding the algorithm error drift.Our proposed method's tracking accuracy and execution time were greatly enhanced using the yoloV5 deep learning detector coupled with DeepSORT and Frame cancellation technique.The results obtained by testing findings on MOT16 datasets showed that the suggested method performed better than other MODTs with YoloV5 and YoloV7 Detectors and MODTs with FCT algorithms.The proposed model achieved an execution time percentage improvement of 61.03%, 60.05%, and 48.31% better than MODT, which has YoloV5 YoloV7 Detectors and MODT with FCT models, respectively.