Gandhimathi (Mathi) Padmanaban
Ph.D., Industrial and Systems Engineering, University of Michigan-Dearborn
On the 2026–27 Academic Job Market · CV · gmathi@umich.edu
I am a researcher in applied computer vision and machine learning. I hold a Ph.D. in Industrial and Systems Engineering and an M.S. in Human Centered Design and Engineering, both from the University of Michigan-Dearborn, where I was advised by Dr. Fred Feng. My dissertation built automated measurement for road safety: estimating how closely vehicles pass cyclists from a single uncalibrated camera, detecting and classifying the vehicles involved, and modelling driver behaviour from vehicle kinematics.
Two problems kept recurring across those projects. The measurements had to hold up when the camera, the platform or the recording site changed. And the system had to be honest about the cases it could not support, because a confident wrong number is worse than no number when a person acts on it. What I found useful in both was checking the output against something outside the model, usually the geometry of how the image was formed.
That is the direction I am taking forward: what class of external checks exists, what makes one informative, and whether the evaluations used to decide a system is fit to deploy measure what people actually rely on it for.
Publications
Detection followed by a fine-tuned ViT over six classes defined by injury risk to cyclists. 89% accuracy at held-out sites with no retraining, and an "unknown" output under low confidence rather than a committed label. In preparation, CVPR 2027. Earlier preprint · Code
Speed-adjusted jerk thresholds jointly optimized with the classifier, because the same jerk magnitude means different things at 20 and 60 mph. 94% accuracy, AUC-ROC 0.971, across 556 trips and seven vehicle models. Submitted, SAE WCX 2027.
Perspective-geometry validation on top of RT-DETR detection and ByteTrack tracking, discarding tracks that are not physically consistent with an overtaking manoeuvre. 98.1% recall with one false positive across 315 events. Submitted, IEEE Transactions on Intelligent Transportation Systems. Poster, TRB Annual Meeting 2026. NSF PAR · Code
Estimates how closely a vehicle passes a cyclist from bounding-box geometry, with no camera calibration. 0.126 m MAE; R² = 0.812 re-fit on the Waymo Open Dataset. In preparation. Details
Transportation Research Board Annual Meeting, 2026. NSF PAR
Ph.D. dissertation, University of Michigan-Dearborn, 2026. Advisor: Dr. Fred Feng.
Whether the standardized cycles used to certify vehicle behaviour represent naturalistic driving. US06 approximates aggressive acceleration but overstates deceleration intensity, and the milder cycles overestimate it as well. WCX SAE World Congress Experience, 2025. SAE Technical Paper 2025-01-8605. DOI
AutomotiveUI '21 Adjunct, ACM, 2021, pp. 142-147. DOI
M.S. thesis, University of Michigan-Dearborn, 2021. Deep Blue