Efficient Object Detection via Adaptive Online Selection of Sensor-Array Elements

Authors: Matthai Philipose

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present empirical evidence over several hundred thousand frames of temperaturegated video from a variety of day-to-day settings that shows an estimated reduction of 50x in power required to detect faces relative to RGB-only processing, at 9% reduction in detection rates. We further break out the benefits of adaptivity, online processing and our optimizations.
Researcher Affiliation Industry Matthai Philipose Microsoft
Pseudocode No The paper describes algorithms and formulations using mathematical equations and descriptive text, but it does not include a clearly labeled pseudocode or algorithm block.
Open Source Code No The paper does not provide any explicit statements about making its source code open or available, nor does it provide a link to a code repository.
Open Datasets No We collected QVGA (320x240) far-infrared video at 10fps with aligned VGA (640x480) RGB frames of daily life in three settings: office , walk and lobby. ... with a total 10hrs of data for each scenario.
Dataset Splits No The paper does not explicitly provide specific training/validation/test dataset splits with percentages, absolute counts, or references to predefined splits. It mentions that "A-O and A-NO are trained on office and lobby data but not walk data" but this does not constitute a formal split description.
Hardware Specification No The paper mentions power consumption estimates for "Proposed silicon implementations of the Viola-Jones algorithm" from other works but does not specify the hardware (e.g., specific GPU/CPU models) used for its own experiments. It refers to "P = 40n J to read a pixel and I = 5n J per instruction" as estimates for calculation, not the hardware used.
Software Dependencies No The paper mentions running "Viola-Jones face detection" but does not provide specific version numbers for any software dependencies or libraries used in their implementation.
Experiment Setup Yes We collected QVGA (320x240) far-infrared video at 10fps with aligned VGA (640x480) RGB frames of daily life in three settings: office , walk and lobby. ... We choose to incur cost sufficient to read and classify a few imager windows per frame: for small n, = n Cp/|W t|. ... We use the Kullback-Leibler divergence D(Pr k Pr0 ). ... we choose i such that Pr(W = 1|gi < i) 0.01 and Pr(W = 1|gi i) 0.3.