Multi-Object Detection and Tracking Using Reptile Search Optimization Algorithm with Deep Learning

Multiple-Object Tracking (MOT) has become more popular because of its commercial and academic potential. Though various techniques were devised for managing this issue, it becomes a challenge factors such as severe object occlusions abrupt appearance changes. presents the optimal outcomes whenever moves uniformly without occlusion in same direction. However, is generally not real scenario, particularly complicated scenes dance events or sporting where greater number players are tracked, moving quickly, varying their speed direction, along with distance position from camera activity they executing. In dynamic scenes, MOT remains main difficulty due to symmetrical shape, structure, size objects. Therefore, study develops new reptile search optimization algorithm deep learning-based multiple detection tracking (RSOADL–MODT) techniques. The presented RSOADL–MODT model intends recognize track objects that exist estimation, tracking, action recognition. It follows series processes, namely detection, classification, tracking. At initial stage, technique applies path-augmented RetinaNet-based (PA–RetinaNet) module, which improves feature extraction process. To improvise network potentiality PA–RetinaNet method, RSOA utilized hyperparameter optimizer. Finally, quasi-recurrent neural (QRNN) classifier exploited classification procedures. A wide-ranging experimental validation process takes place on DanceTrack MOT17 datasets examining effectual algorithm. simulation values confirmed enhancements method over other DL approaches.