Anti-occlusion pedestrian tracking algorithm based on location prediction and deep feature rematch

Aiming to the problem of pedestrian tracking with frequent or long-term occlusion in complex scenes,an anti-occlusion pedestrian tracking algorithm based on location prediction and deep feature rematch is proposed.Firstly,the occlusion judgment is realized by extracting and utilizing deep feature of pedestrian’s appearance,and then the scale adaptive kernelized correlation filter is introduced to implement pedestrian tracking without occlusion.Secondly,Karman filter is introduced to predict the location of occluded pedestrian position.Finally,the deep feature is used to the rematch of pedestrian in the reappearance process.Simulation experiment and analysis show that the proposed algorithm can effectively detect and rematch pedestrian under the condition of frequent or long-term occlusion.

Bounding box based on probability density for the pedestrian tracking

This paper proposes a pedestrian tracking approach using bounding box based on probability densities.It is generally a difficult task to track features like corner points in outdoor images due to complex environment.To solve this problem,the feature points are projected along X and Y direction separately,and a histogram is constructed for each projection,with horizontal axis as positions and vertical axis as the number of feature points that lie on each position.Finally,the vertical axis is normalized for expression as probability.After histogram is constructed,the probability of each feature point is checked with a threshold.A feature point will be ignored if its probability is lower than a threshold,while the remaining feature points are grouped,based on which a bounding box is made.Kanade-Lucas Tomasi(KLT)algorithm is adopted as the tracking algorithm because it is able to track local features in images robustly.The efficiency of the tracking results using this method is verified in real environment test.

Indoor Pedestrian Tracking with Sparse RSS Fingerprints

Indoor pedestrian localization is of great importance for diverse mobile applications. Many indoor localization approaches have been proposed; among them, Radio Signal Strength （RSS）-based approaches have the advantage of existing infrastructures and avoid the cost of infrastructure deployment. However, the RSS-based localization approaches suffer from poor localization accuracy when the RSS fingerprints are sparse, as illustrated by actual experiments in this study. Here, we propose a novel indoor pedestrian tracking approach for smartphone users; this approach provides a high localization accuracy when the RSS fingerprints are sparse. Besides using the RSS fingerprints, this approach also utilizes the inertial sensor readings on smartphones. This approach has two components： （i） dead-reckoning subsystem that counts the number of walking steps with off-the-shelf inertial sensor readings on smartphones and （ii） particle filtering that computes the locations with only sparse RSS readings. The proposed approach is implemented on Android-based smartphones. Extensive experiments are carried out in both srnafl and large testbeds, The evaluation results show that the tracking approach can achieve a high accuracy of 5 m （up to 95%） in indoor environments with only sparse RSS fingerprints.

A novel method for tracking pedestrians from real-time video

This novel method of Pedestrian Tracking using Support Vector (PTSV) proposed for a video surveillance instrument combines the Support Vector Machine (SVM) classifier into an optic-flow based tracker. The traditional method using optical flow tracks objects by minimizing an intensity difference function between successive frames, while PTSV tracks objects by maximizing the SVM classification score. As the SVM classifier for object and non-object is pre-trained, there is need only to classify an image block as object or non-ob-ject without having to compare the pixel region of the tracked object in the previous frame. To account for large motions between successive frames we build pyramids from the support vectors and use a coarse-to-fine scan in the classification stage. To accelerate the training of SVM, a Sequential Minimal Optimization Method (SMO) is adopted. The results of using a kernel-PTSV for pedestrian tracking from real time video are shown at the end. Comparative experimental results showed that PTSV improves the reliability of tracking compared to that of traditional tracking method using optical flow.

A robust system for real-time pedestrian detection and tracking

A real-time pedestrian detection and tracking system using a single video camera was developed to monitor pedestrians. This system contained six modules: video flow capture, pre-processing, movement detection, shadow removal, tracking, and object classification. The Gaussian mixture model was utilized to extract the moving object from an image sequence segmented by the mean-shift technique in the pre-processing module. Shadow removal was used to alleviate the negative impact of the shadow to the detected objects. A model-free method was adopted to identify pedestrians. The maximum and minimum integration methods were developed to integrate multiple cues into the mean-shift algorithm and the initial tracking iteration with the competent integrated probability distribution map for object tracking. A simple but effective algorithm was proposed to handle full occlusion cases. The system was tested using real traffic videos from different sites. The results of the test confirm that the system is reliable and has an overall accuracy of over 85%.

CGTracker:Center Graph Network for One-Stage Multi-Pedestrian-Object Detection and Tracking

Most current online multi-object tracking(MOT)methods include two steps:object detection and data association,where the data association step relies on both object feature extraction and affinity computation.This often leads to additional computation cost,and degrades the efficiency of MOT methods.In this paper,we combine the object detection and data association module in a unified framework,while getting rid of the extra feature extraction process,to achieve a better speed-accuracy trade-off for MOT.Considering that a pedestrian is the most common object category in real-world scenes and has particularity characteristics in objects relationship and motion pattern,we present a novel yet efficient one-stage pedestrian detection and tracking method,named CGTracker.In particular,CGTracker detects the pedestrian target as the center point of the object,and directly extracts the object features from the feature representation of the object center point,which is used to predict the axis-aligned bounding box.Meanwhile,the detected pedestrians are constructed as an object graph to facilitate the multi-object association process,where the semantic features,displacement information and relative position relationship of the targets between two adjacent frames are used to perform the reliable online tracking.CGTracker achieves the multiple object tracking accuracy(MOTA)of 69.3%and 65.3%at 9 FPS on MOT17 and MOT20,respectively.Extensive experimental results under widely-used evaluation metrics demonstrate that our method is one of the best techniques on the leader board for the MOT17 and MOT20 challenges at the time of submission of this work.