Laser & Optoelectronics Progress, Volume. 62, Issue 4, 0400004(2025)
Review of Event Camera-Based Target Detection and Tracking Algorithms
Figures & Tables(16)
Fig. 5. Processing steps of graph neural network. (a) Event stream; (b) subsampling; (c) graph generation; (d) GNN; (e) prediction
Fig. 9. Algorithmic framework for real-time optical flow estimation based on pulsed neural network[50]
Fig. 10. Framework of mixed frame-/event-driven pedestrian detection[55]. (a) Tracker framework; (b) frame attention module
Fig. 11. Framework for event-based object motion estimation for target tracking methods[60]
Fig. 12. A framework for target tracking algorithms with multimodal fusion[63]. (a) Overview of cross-domain feature integrator; (b) edge-attention block; (c) cross-domain modulation and selection block; (d) Bbox regressor; (e) classifier
Table 1. Comparison of the characteristics and uses of the three sensors
View tableTable 1. Comparison of the characteristics and uses of the three sensors
Type Advantage Disadvantage Application scenario DVS It only outputs event information and is able to maintain the advantages of low latency， low power consumption and high dynamic range Unable to output any grayscale information， poor visibility It is suitable for situations with low requirements for visualization and high requirements for low latency， such as tracking， counting or motion monitoring of high-speed moving objects， real-time positioning ATIS It is capable of outputting event information and gray scale information of pixels where light intensity changes occur， with some visibility It is prone to exposure anomalies and information loss when the ambient brightness changes slowly It is suitable for frequent changes in environmental brightness and the existence of high-speed movement occasions， mainly used for real-time monitoring of products in industrial manufacturing DAVIS It can combine event information and grayscale information with good visibility， high temporal resolution， and has the ability to obtain detailed information about movements It can be affected by defects such as redundancy of information in APS cameras， low temporal resolution， and low dynamic range It is suitable for occasions with high visualization requirements and small dynamic range， such as target recognition， target detection， tracking and localization of targets， especially in the field of robotics and unmanned vehicles
Table 2. Object detection algorithms based on event camera
View tableTable 2. Object detection algorithms based on event camera
Type Category Reference Year/source Contribution Pure stream of events Angle detection ［ 18 ］2015/Neural Networks Transforming the traditional problem of corner point detection into a local velocity estimation problem ［ 19 ］2015/IEEE Consideration of spatial correlation between data ［ 20 ］2017/IEEE Realization of the estimation of the velocity of a corner point ［ 21 ］2023/ICVR Design of an Enhanced Representation of Time Surface Events ［ 22 ］2023/IEEE Proposing a novel asynchronous corner point detector Arc* Linear Detection ［ 23 ］2016/BMVC Development of an event-based Hough transform method ［ 24 ］2018/IEEE Proposing weighted least squares fitting method based on iterative events Pure stream of events Deep learning ［ 25 ］2021/IEEE Proposing fast graph construction method based on radius search method ［ 26 ］2022/ IEEE Design of an asynchronous event-based graph neural networks （AEGNNs） ［ 27 ］2021/Neural Networks Proposing fully Convolutional Recurrent Neural Network Architecture ［ 28 ］2023/arXiv Proposing a dual-memory aggregation network （DMANet） ［ 29 ］2023/Peking University Proposing pulsed streaming target detection method based on asynchronous spatio-temporal memory network （ASTMNet） ［ 30 ］2023/ICRA Proposing a lightweight spiking neural network that can separate events based on the speed of the corresponding objects Combination of event streams and traditional frames Deep learning ［ 31 ］2016/ IEEE Combining CNN and Particle Filtering ［ 32 ］2018/ IEEE Training CNNs on pseudo-labels ［ 33 ］2018/ IEEE Applying existing deep learning methods to the iCub robotics platform ［ 34 -35 ］2019/Frontiers in Neurorobotics Using three different event stream encoding methods and two fusion methods ［ 36 ］2019/IEEE Proposing a confidence fusion method based on CNN detection results ［ 37 ］2022/IEEE Applying Asynet to maritime incident datasets for target recognition ［ 38 ］2023/IEEE Exploring a multistage design approach and proposing Recurrent Vision Transformers for object detection ［ 39 ］2023/Chinese Academy of Sciences Proposing an adaptive temporal resolution method for visualizing event information
Table 3. Event camera-based target tracking algorithm
View tableTable 3. Event camera-based target tracking algorithm
Type Category Reference Year/source Contribution Pure stream of events Angle detection ［ 40 ］2018/IEEE Establishing correspondence between asynchronous detections in the event stream in real time ［ 41 ］2018/IEEE Establishing local correspondences between corner events Pure stream of events ［ 42 ］2021/Zhejiang University of Technology Adding speed constraints to the ACE tracker Clustering ［ 43 ］2012/IEEE Gaussian mixture modeling to form event clusters for object characterization ［ 44 ］2017/IEEE Combination of stereo matching and object tracking ［ 45 ］2018/IEEE mean-drift-clustering-based approach for event labeling ［ 46 ］2021/IEEE Applying SVM to event cameras，combining local and global sliding window search Optical flow estimation ［ 47 ］2012/Neural Networks Replacing the gray value gradient by comparing the instantaneous activity values of neighboring active pixels ［ 48 ］2020/IEEE Proposing an unsupervised learning algorithm for optical flow estimation of sparse event data ［ 49 ］2023/IEEE Proposing incremental full flow estimation from sparse normal flow based measurements ［ 50 ］2023/IEEE Proposing a real-time estimation algorithm for neuromorphic optical flow based on impulse neural networks Deep learning ［ 51 ］2019/Frontiers in Neurorobotics Extraction of features after coding the event stream based on rate coding ［ 52 ］2022/Advances in Neural Information Processing Systems Downsampling strategy to mine key events and embed the irregular spatio-temporal information of key events into a high-dimensional feature space Combination of event streams and traditional frames Feature tracking ［ 53 ］2016/IEEE Proposing a feature trajectory tracking algorithm that combines event streams and grayscale maps ［ 54 ］2017/IEEE An expectation maximization （EM） algorithm is used to quantify the probabilistic correlation between feature point sets and event data ［ 55 ］2023/IEEE A novel frame-attention module is introduced Correlation filtering ［ 42 ］2021/Zhejiang University of Technology Kernel-correlation filtering is introduced Particle filtering ［ 56 ］2017/IEEE Multi-hypothesis filtering technology is introduced Clustering ［ 57 ］2018/CHREOC Clustering based on spatial relationships ［ 58 ］2019/ICCS Hierarchical clustering ［ 59 ］2021/ICCC Alignment and foreground enhancement models are introduced Deep learning ［ 60 ］2020/AAAI Designing a temporal surface representation of TSLTD ［ 61 ］2021/Dalian University of Technology Channel Attention Mechanism added to Feature Fusion Module is introduced ［ 62 ］2022/IEEE STNET is proposed to dynamically extract and fuse information from the temporal and spatial domains ［ 63 ］2021/IEEE Discretizing asynchronous events into time slices accumulated in conventional frames and designing a cross-domain feature integrator ［ 64 ］2023/IEEE Design of an event-guided cross-modal alignment and crossover based module ［ 65 ］2023/arXiv Cross-modal hierarchical knowledge distillation scheme is introduced ［ 66 ］2024/IEEE Development of a cross-modal converter for bimodal information interaction ［ 67 ］2022/Sensors Adaptive strategy to adjust the spatial and temporal domain of event data for event frame reconstruction
Table 4. Comparison of event camera-based visual tracking datasets
View tableTable 4. Comparison of event camera-based visual tracking datasets
Dataset Year Videos Frames Class Attributes Resolution Aim Public VOT-DVS 2016 60 240×180 Eval √ TD-DVS 2016 77 240×180 Eval √ Ulster 2016 1 9000 240×180 Eval × EED 2018 7 234 240×180 Eval √ FE108 2021 108 208672 21 4 346×260 Train&Eval √ VisEvent 2021 820 371127 17 346×260 Train&Eval √
Tools
Get Citation
Copy Citation Text
Jiayu Qiu, Yasheng Zhang, Yuqiang Fang, Pengju Li, Kaiyuan Zheng. Review of Event Camera-Based Target Detection and Tracking Algorithms[J]. Laser & Optoelectronics Progress, 2025, 62(4): 0400004
Paper Information
Category: Reviews
Received: Apr. 10, 2024
Accepted: Jun. 17, 2024
Published Online: Feb. 10, 2025
The Author Email:
CSTR:32186.14.LOP241073
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20