Our Elixir is a system that enhances video analytics performance from multiple IoT cameras by adjusting camera settings through Multi-Objective Reinforcement Learning (MORL). In tests, Elixir significantly outperformed default settings and time-sharing methods, improving detection rates for cars, faces, persons, and license plates, addressing challenges in multi-analytical unit settings.
At the 9th International Conference on Internet of Things: Systems, Management and Security (IOTSMS 2022), we presented a novel reinforcement-learning system called APT, which dynamically adjusts camera parameters remotely via 5G networks. Experiments showed a significant accuracy improvement in video analytics, particularly a 42% enhancement in object detection, demonstrating APT's applicability across various tasks.
The paper presented at the 20th ACM Conference on Embedded Networked Sensor Systems (SenSys 2022) discusses a framework called CamTuner, which dynamically adjusts non-automated (NAUTO) camera parameters to enhance the accuracy of analytics units in video analytics pipelines (VAP). The researchers highlight that environmental changes can severely impact the accuracy of insights derived from surveillance cameras. By utilizing SARSA reinforcement learning and incorporating an analytics quality estimator alongside a virtual camera to expedite training, CamTuner improves detection accuracy significantly—detecting 15.9% more individuals and increasing vehicle recognition rates in diverse scenarios—thereby enabling advanced applications like automatic vehicle collision prediction.
We investigate ways to leverage uncertainty in face images to improve the quality of the face clusters. We observe that popular clustering algorithms do not produce better quality clusters when clustering probabilistic face represen- tations that implicitly model uncertainty – these algorithms predict up to 9.6X more clusters than the ground truth for the IJB-A benchmark. We empirically analyze the causes for this unexpected behavior and identify excessive false- positives and false-negatives (when comparing face-pairs) as the main reasons for poor quality clustering. Based on this insight, we propose an uncertainty-aware clustering al- gorithm, UAC, which explicitly leverages uncertainty infor- mation during clustering to decide when a pair of faces are similar or when a predicted cluster should be discarded. UAC considers (a) uncertainty of faces in face-pairs, (b) bins face-pairs into different categories based on an un- certainty threshold, (c) intelligently varies the similarity threshold during clustering to reduce false-negatives and false-positives, and (d) discards predicted clusters that ex- hibit a high measure of uncertainty. Extensive experimen- tal results on several popular benchmarks and comparisons with state-of-the-art clustering methods show that UAC pro- duces significantly better clusters by leveraging uncertainty in face images – predicted number of clusters is up to 0.18X more of the ground truth for the IJB-A benchmark.
Giuseppe Coviello 