Abstract
While most filtering approaches based on random finite sets have focused on improving performance, in this paper, we argue that computation times are very important in order to enable real-time applications such as pedestrian detection. Towards this goal, this paper investigates the use of OpenCL to accelerate the computation of random finite set-based Bayesian filtering in a heterogeneous system. In detail, we developed an efficient and fully-functional pedestrian-tracking system implementation, which can run under real-time constraints, meanwhile offering decent tracking accuracy. An extensive evaluation analysis was carried out to ensure the fulfillment of sufficient accuracy requirements. This was followed by extensive profiling analysis to spot the potential bottlenecks in terms of execution performance, which were then targeted to come up with an OpenCL accelerated application. Video-throughput improvements from roughly 15 fps to 100 fps (6×) were observed on average while processing typical MOT benchmark videos. Moreover, the worst-case frame processing yielded an 18× advantage from nearly 2 fps to 36 fps, thereby comfortably meeting the real-time constraints. Our implementation is released as open-source code.
Highlights
Current success in the practical implementations of random finite set (RFS) filters has made it clear that RFS-based approaches are going to play a key role in the multisensor data fusion
The RFS-based filtering enables the use of the optimal Bayesian estimation framework for multi-target tracking scenarios by introducing the concepts of a multi-target state/measurement expressed via random finite sets
After extensive simulation analysis (Section 4), we found the Gaussian mixture (GM)-probability hypothesis density (PHD) filter accuracy to be inadequate in dealing with general pedestrian tracking scenarios where the pedestrians deviated from linear/Gaussian motion characteristics
Summary
Current success in the practical implementations of random finite set (RFS) filters has made it clear that RFS-based approaches are going to play a key role in the multisensor data fusion. We investigate the problem of applying RFS filtering approaches to a heterogeneous platform, aiming to provide some insights on how to improve the RFS filtering running times by the heterogeneous system For this purpose, we chose recently-proposed RFS-based filtering techniques i.e., PHD, labeled multi-Bernoulli (LMB) filters, to tackle the underlying multi-target tracking problem. The accuracy should be guaranteed while deploying OpenCL acceleration To this end, the tracking algorithms of this paper were developed in a highly modular system design approach and practically implemented in tracking pedestrians of a video.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
