Abstract:Multi-similar appearance and occlusion problem are still the main difficult issues in the field of multi-target tracking. Many practical multi-object tracking (MOT) algorithms based on detectors widely used in industry only rely on the detector's Intersection over Union (IOU) to solve most tracking problems. In this paper, we propose a MOT strategy based on a semi-online mechanism, which includes IOU matching strategy, trajectory exploration strategy and backtracking mechanism. Meanwhile, the integrated motion model and the enhanced appearance model are also presented. According to the tracking temporal window, our strategy will find multiple pairs of detection frames by matching the largest IOU first, and then it leverages Kalman filter to establish a motion model for exploring various tracking trajectory. At last, it calculates the reliability of different tracking trajectories and executes regret strategy we proposed to generate optimized tracking trajectory. Our algorithm focuses on transforming the tracking issues into how to generate robust fragments of tracking trajectory,and how to fix the back-tracking errors within the temporal window according to chronological order of tracking context. Our method is naturally suitable for solving the difficult issues such as disappearance, detection noise, and target occlusion. We evaluate our algorithms on the datasets of MOT2015 and MOT2017. Compared with baseline algorithm, our method reduces 16 IDS on each video sequence on average, and the multi-target tracking accuracy has been improved by 4.01%. Our algorithm model is very versatile and efficient for the algorithms of short trajectory stitching. Keywords—Semi-online, MOT, Regret Strategy, IOU Mask, Tracklet Association, Low IDS.

An Optimized Target Tracking Algorithm of Online Boosting

Rapid visual tracking with modified on-line boosting and template matching

Part-Based Online Tracking with Geometry Constraint and Attention Selection

An Experimental Comparison Of Online Object Tracking Algorithmsu

Feature-Based On-Line Object Tracking Combining Both Keypoints and Quasi-Keypoints Matching.

Online learning affinity measure with CovBoost for multi-target tracking

An efficient AdaBoost tracking algorithm based on the particle framework

An online system for multiple interacting targets tracking

Scale and rotation invariant feature-based object tracking via modified on-line boosting

Adaboost Object Tracking Algorithm

An anti-occlusion optimization algorithm for multiple pedestrian tracking

Object tracking algorithm based on system ensemble

A new robust visual tracking algorithm based on Transfer Adaptive Boosting

An Online System For Multiple Interacting Targets Tracking: Fusion Of Laser And Vision, Tracking And Learning

Co-training algorithm based on on-line boosting for vehicle tracking

A Novel Strategy for Multi-Objective Tracking Framework based on Semi-online Mechanism

Online multi-object tracking with efficient track drift and fragmentation handling

An Improved Algorithm for Target Tracking Based on Particle Filter

Online Transfer Boosting for object tracking

Accurate target tracking based on off-tracking and crossed occlusione

Occlusion-Aware Real-Time Object Tracking