Comparing event selection sensitivity in video trigger systems
Readout and Operating-Point Sensitivity in Event-Stream Candidate Selection
Computer Vision and Pattern Recognition
Summary
Detecting important moments in video streams depends on correctly choosing when to record. The authors studied how different methods and settings affect capturing real events versus false alarms. They tested several neural network models on real video data with many people and found differences in accuracy and false triggers depending on the model and calibration. No single method clearly outperformed others in all cases, and settings like viewing duration also changed outcomes. The study focused on offline analysis and does not show how well these methods perform in everyday, continuous use.
What this means in practice
- •For video analytics developers: Optimize event-triggered video capture systems by selecting encoders and calibration settings that balance target detection and false alarms based on offline evaluations.
- •For wearable device engineers: Design candidate event detectors with lightweight models and calibrated thresholds to improve event selection performance in energy-constrained devices.
Tested on one dataset.
Authors
Yifan Shuai
Abstract
Event-triggered recording needs a useful combination of target capture and false triggers, yet representation comparisons can change with readout and threshold calibration. We evaluate real event accumulations from 520 DailyDVS-200 clips and 47 participants using participant-disjoint training, validation, and testing. The original study varies frozen encoder scale, accumulation duration, and temporal view count; a post-review extension adds two supervised-pretrained convolutional encoders, a 61,586-parameter trained CNN with three seeds, ten repeated participant partitions, and calibration sensitivity. At 50 ms/eight views under the primary handshake-or-fall policy, DINO-S CLS captures 101/132 targets with 40/388 false triggers, versus 78/132 and 35/388 for patch mean. DINO-B CLS has lower selected-threshold capture but higher fold-wise low-FPR partial AUC (0.746 versus 0.732). MobileNetV3-Small captures 97/132 with 53/388 false triggers, demonstrating why lightweight baselines and actual operating points matter. Across ten dependent repartitions, mean normalized partial AUC is 0.738 for DINO-B CLS and 0.674 for MobileNet. The original LeJEPA S-to-B capture difference is 8.3 points, but its multiplicity-adjusted interval includes zero. Duration and view-count effects remain policy dependent. We find no established JEPA-specific advantage. The evidence concerns offline action-defined candidate selection on previously used development data; it does not establish personal recording value, continuous false alarms per hour, or eyewear energy.