Autonomous UAV Navigation for Search and Rescue Missions Using Computer Vision and Convolutional Neural Networks

Bojan \v{S}ekoranja; Branimir \'Caran; Luka \v{S}iktar; Marko \v{S}vaco

REVIEW 5 major objections 6 minor 13 references

Autonomous UAV Navigation for Search and Rescue Missions Using Computer Vision and Convolutional Neural Networks

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that a small drone can search for a known person, recognize their face, and follow them at a safe distance using only a single camera and neural networks.

desk verdict A credible real-time UAV detect-recognize-track pipeline, but the load-bearing distance estimator is never validated against ground truth, so the 2 m tracking claim is unsupported. read the letter →

arxiv 2507.18160 v1 pith:T5IG5LUL submitted 2025-07-24 cs.RO

Luka Šiktar , Branimir Ćaran , Bojan Šekoranja , Marko Švaco This is my paper · ORCID

classification cs.RO

keywords UAVsearchandrescuecomputervisionconvolutionalneuralnetworkYOLOv11facerecognitionposeestimationPDcontrol

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is trying to show that a small quadcopter can perform a complete search-and-rescue subtask on its own: detect a person, recognize whether that person is the one being sought, and then follow them while keeping a set distance. The system runs on a DJI Tello UAV with video streamed to a Jetson Xavier NX board, using YOLOv11 to spot people, a dlib face-recognition network to identify the target, and YOLOv11-pose to estimate body keypoints. Distance to the person is inferred from the pixel length of the shoulder-to-hip segment through a calibrated quadratic formula, and three PD controllers use that distance to drive forward, vertical, and yaw motion. If correct, this provides a compact proof of concept for autonomous UAV assistance in search and rescue, with the perception happening off the ground station.

What carries the argument

The load-bearing mechanism is the shoulder-to-hip keypoint segment extracted by YOLOv11-pose. Its pixel length $x$ is mapped to a physical distance estimate $y = k_1 x^2 + k_2 x + k_3$ in centimeters, calibrated experimentally from 0 to 6 meters on average European body heights. Because shoulders and hips remain visible from many orientations, this single monocular cue lets the UAV track a person whose face is not visible, and it supplies the error signal to three PD controllers governing X translation, Z translation, and yaw. The system identification step provides the linear dynamic models for each axis that make the PD gains appropriate.

What would settle it

Place people of known different heights, say 160 cm and 190 cm, at measured distances from the UAV, hold posture fixed, and compare the system's reported distance with a tape measure or a distance sensor; if the error grows with height or with a 20-degree lean while true distance is constant, the distance signal is biased.

Watch

Extended reading notes

Core claim

The central discovery is that the full detect-recognize-track loop can be closed on a small UAV using only a monocular camera. YOLOv11 filters its detections to people, dlib's face embedding compares each detected face against a stored template, and YOLOv11-pose returns 17 body keypoints. The shoulder-to-hip pixel distance is converted to physical distance with a fitted quadratic, and that distance plus the shoulder midpoint's offset from the image center feed three PD controllers for forward/backward, up/down, and yaw. The PD gains come from data-driven system identification of the Tello's dynamics using IMU telemetry. In tests the loop ran in real time, with face recognition only in the search phase and tracking following the body rather than the face.

Load-bearing premise

The load-bearing premise is that the pixel length of a person's shoulder-to-hip segment is a stable ruler for physical distance, but the quadratic mapping is calibrated on average European heights and does not account for individual height or posture, so a shorter, taller, or leaning person produces a biased distance that the PD controllers treat as true.

Editorial extensions

If this is right

The UAV can follow a person from angles where the face is hidden, as long as shoulders and hips are visible.
An operator can initialize the system without a stored photo by centering the person and saving a template mid-flight.
The loop's timing budget is set by control at 15 Hz and pose at 15 Hz, which is adequate for slow human motion.
If a Wi-Fi glitch interrupts tracking, the system can re-run face recognition against stored templates to reacquire the target.
The same architecture can be ported to a larger UAV with better cameras and global positioning, as the authors state as next steps.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The distance estimate has not been checked against ground truth; a direct range test with people of different heights and body angles would show whether the controllers are holding physical distance or merely keeping a pixel size constant.
Because the shoulder-to-hip calibration is tied to average European body dimensions, the same system would need re-calibration for populations with different height distributions.
The pose model runs at half the frame rate of detection, so the limiting factor for tracking smoothness is likely the pose/control chain rather than the person detector; raising pose estimation to the same rate as detection is a testable improvement.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: this is a real, working proof of concept, and the stress-test concern is correct. The authors built a ROS2 pipeline on a DJI Tello with a Jetson Xavier NX: YOLOv11 detects people, dlib recognizes faces, YOLOv11-pose extracts body keypoints, and three PD controllers keep the chosen individual centered and at a set distance. Getting that loop to run in real time on a small UAV is an actual engineering achievement, and the GitHub repo with videos is a plus.

What is new is the use of shoulder-to-hip keypoints, rather than the face, as the distance reference, with a fitted quadratic mapping from pixel span to centimeters. That is a modest extension of prior face-tracking work, but it is genuinely useful for scenarios where the face is not visible, and the system-level integration of YOLOv11 + dlib + YOLOv11-pose is not in the cited literature.

The soft spots are real. Equation (1) is the single distance signal feeding all three controllers. The authors give no fitted k1, k2, k3, no calibration protocol, and no ground-truth comparison between estimated and measured distance. The claim that the drone maintains two meters is therefore not substantiated. The calibration assumes an average European height, which will bias the estimate for shorter and taller people, and the authors themselves note that leaning distorts the shoulder-to-hip span. None of this is fatal for a proof of concept, but it means the central quantitative claim is unvalidated.

Two smaller issues: the abstract says 14 known individuals while Section 4 says 11, and the paper reports no precision/recall or error metrics for detection, recognition, or tracking. For an engineering evaluation, that is the main gap.

Who benefits: anyone working on UAV person-following or search-and-rescue systems will get a clear picture of a functioning baseline. The paper deserves peer review rather than desk rejection, because the system is real, the code appears available, and the missing pieces are fixable. My recommendation: send it out—workshop or applied venue—with a clear request for major revisions: report the fitted coefficients, validate Eq. (1) against ground truth or known-height subjects, and add quantitative tracking errors. I would not cite the 2 m tracking claim until that is done, but I would cite the paper as related work on pose-based distance control.

Referee Report

5 major / 6 minor

Summary. The paper presents an autonomous UAV subsystem for search and rescue built on a DJI Tello equipped with a Jetson Xavier NX and running ROS2. The system uses YOLOv11 for people detection, dlib for face recognition, and YOLOv11-pose for body keypoint extraction, and it computes a monocular distance estimate from the shoulder-to-hip pixel span. Three PD controllers, tuned through data-driven system identification, regulate forward/backward, vertical, and yaw motion, with a target tracking distance set to 2 m. The authors report qualitative real-time indoor and outdoor experiments and state that the complete solution, videos, and evaluations are available on GitHub.

Significance. If properly validated, the system would be a useful proof of concept for a compact, low-cost UAV performing the full detect-recognize-track loop in real time with a monocular camera, which is relevant to search and rescue operations. The paper's strengths are the integration of modern CNNs in a ROS2 pipeline, the use of data-driven system identification for controller tuning, and the public availability of the implementation and test videos. However, the central quantitative claim of maintaining a specified 2 m distance is not currently supported: the distance estimator is not validated against ground truth, the controller gains and identified model parameters are not reported, and the experimental section contains no quantitative accuracy metrics for detection, recognition, or tracking.

major comments (5)

[Section 4, Eq. (1)] The claim that the UAV maintains a specified distance (2 m, Section 3.5) is load-bearing and currently unsupported. Equation (1) is the sole distance signal for all three PD controllers, yet the paper reports no fitted coefficients k1, k2, and k3, no calibration protocol, and no comparison against measured distances. Add the calibration procedure, the coefficient values, and error statistics (e.g., RMSE or MAE against a measured range over 0–6 m) to substantiate the metric-distance claim.
[Section 4] The experimental evaluation is entirely qualitative: there are no precision/recall or mAP numbers for people detection, no recognition accuracy or false-accept/reject rates for face recognition, and no tracking error or distance-hold statistics. The statement that 'all the proposed deep learning methods demonstrated excellent performance' is not measurable. Report quantitative metrics with defined success criteria for detection, recognition, and tracking.
[Abstract vs. Section 4] The number of test individuals is inconsistent: the abstract states that experiments were conducted on 14 known individuals, while Section 4 states that the system was tested on 11 individuals. Clarify which number is correct and report the number of trials, the environments used, and any exclusion criteria.
[Section 3.1] The system identification and PD tuning are described only in words. The identified transfer functions or state-space models, the PD gains, and the Simulink closed-loop validation (nominally shown in Fig. 2) are not provided. Without these values, the 'data-driven system identification' contribution cannot be assessed. Report the identified model parameters and gains, along with a model-fit metric (e.g., fit percentage) and a closed-loop performance measure.
[Section 3.4, Eq. (1)] Equation (1) assumes a stable shoulder-to-hip reference length and average European heights (180 cm male, 171 cm female). For individuals of different height or body proportions, and for postures such as leaning, the distance estimate is systematically biased, and the controllers treat this biased estimate as truth. The paper acknowledges leaning as a limitation but does not quantify its effect or validate the estimator across different subjects and poses. This is load-bearing for the 'maintains a specified distance' claim and should be addressed with multi-subject, ground-truth validation.

minor comments (6)

[Section 3.3] The paper uses dlib's face recognition model but cites reference [11], which describes FaceNet; dlib's model is a different ResNet-based architecture, so the citation should be to dlib's documentation or to an appropriate source for that model.
[Sections 3.2 and 4] The processing rates are inconsistent: Section 3.2 states real-time detection at 15 Hz, while Section 4 states that YOLOv11 processes frames at 30 Hz and YOLOv11-pose at 15 Hz. Reconcile these numbers.
[Fig. 2] The graph labels 'graph1' and 'graph2' are not descriptive; relabel the subplots as input signal and output velocity for clarity.
[Section 3.1] The terminology for the controlled axes is confusing: the text refers to a 'roll/X-axis PD controller' and later to 'yaw/Z-axis linear movement.' Use consistent notation for translational axes (X, Z) and rotational axes (yaw).
[Section 4] The phrase 'losing traction and failing to maintain tracking procedure' appears to be a typo for 'losing track'; please correct the wording.
[References] Several references are incomplete or inconsistent with a journal style, such as [5] lacking page numbers and [10] lacking volume and page details; please check all entries against the target journal's reference format.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the distance calibration and system identification are openly data-driven; the main weakness is missing ground-truth validation, not circularity.

full rationale

The paper's derivation chain is not circular. Equation (1) defines the relative distance estimate as a quadratic function of shoulder-to-hip pixel distance, with coefficients that the paper states were 'determined experimentally by deriving the polynomial equation for measured distances, accurately calibrated in the range 0 to 6 meters' (Section 3.4). This is an explicit calibration from measured data rather than a prediction derived from the same quantity it claims to predict. The three PD controllers are designed using system identification from recorded IMU command/response data and tuned in Simulink (Section 3.1), again an openly data-driven procedure rather than a hidden reuse of the output as an input. The final claim of tracking at a specified distance of 2 m does depend on the pixel-to-metric mapping, and the paper reports no independent ground-truth distance error statistics; it also acknowledges that leaning can distort the shoulder-to-hip distance. This is a real validation and robustness gap, but it is not circular reduction: the distance formula was not fitted to the tracking outcome, and the controller evaluation is not claimed to be independent of the identified model. There is also no load-bearing self-citation or imported uniqueness theorem; the cited FaceNet threshold and YOLOv11 pretrained models are external, machine-checkable or standard resources. Therefore, no specific step reduces to its own input by construction.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The paper rests on openly stated modeling choices: linear dynamics for the UAV, a fixed shoulder-to-hip reference length, average European height, and standard pretrained models. None of these is an invented entity; all are domain assumptions with no quantitative validation in the paper.

free parameters (2)

k1, k2, k3 in distance model Eq. (1) = not reported
Quadratic mapping from shoulder-to-hip pixel distance to centimeters; calibrated from measured distances in the 0 to 6 meter range, but the fitted coefficients are not given in the paper.
PD controller gains for X, Z, and yaw axes = not reported
Tuned through Matlab System Identification Toolbox and Simulink; actual gains are omitted, so the published paper alone cannot reproduce the control behavior.

assumptions (5)

domain assumption The UAV dynamics for forward/backward, up/down, and yaw can be described by linear differential equations.
Section 3.1 states this assumption to simplify system identification.
domain assumption The shoulder-to-hip pixel distance is invariant to the person's orientation relative to the camera.
Section 3.4 motivates the body-keypoint distance approach with this claim; it is not validated with quantitative pose-variation tests.
domain assumption Average European male height (180 cm) and female height (171 cm) are valid reference dimensions for tracked people.
Section 3.4 uses these heights to map pixel distances to physical distances.
domain assumption The dlib face recognition threshold of 0.6 Euclidean distance correctly separates same and different identities on low-resolution UAV imagery.
Section 3.3 adopts the threshold from FaceNet [11] without validation on the UAV camera domain.
domain assumption YOLOv11 and YOLOv11-pose pretrained on COCO provide sufficient person and keypoint accuracy on 720p UAV video.
Section 4 asserts excellent performance but provides no quantitative detection or pose accuracy numbers.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Autonomous UAV Navigation for Search and Rescue Missions Using Computer Vision and Convolutional Neural Networks." pith.science (2026). https://pith.science/paper/T5IG5LUL

@misc{pith2026250718160,
  author       = {Pith},
  title        = {Pith review of: Autonomous UAV Navigation for Search and Rescue Missions Using Computer Vision and Convolutional Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T5IG5LUL}},
  note         = {Machine review of arXiv:2507.18160}
}

read the original abstract

In this paper, we present a subsystem, using Unmanned Aerial Vehicles (UAV), for search and rescue missions, focusing on people detection, face recognition and tracking of identified individuals. The proposed solution integrates a UAV with ROS2 framework, that utilizes multiple convolutional neural networks (CNN) for search missions. System identification and PD controller deployment are performed for autonomous UAV navigation. The ROS2 environment utilizes the YOLOv11 and YOLOv11-pose CNNs for tracking purposes, and the dlib library CNN for face recognition. The system detects a specific individual, performs face recognition and starts tracking. If the individual is not yet known, the UAV operator can manually locate the person, save their facial image and immediately initiate the tracking process. The tracking process relies on specific keypoints identified on the human body using the YOLOv11-pose CNN model. These keypoints are used to track a specific individual and maintain a safe distance. To enhance accurate tracking, system identification is performed, based on measurement data from the UAVs IMU. The identified system parameters are used to design PD controllers that utilize YOLOv11-pose to estimate the distance between the UAVs camera and the identified individual. The initial experiments, conducted on 14 known individuals, demonstrated that the proposed subsystem can be successfully used in real time. The next step involves implementing the system on a large experimental UAV for field use and integrating autonomous navigation with GPS-guided control for rescue operations planning.

Figures

Figures reproduced from arXiv: 2507.18160 by the authors.

**Figure 4.** Fig.4 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

[1]

Use of Unmanned Aerial Vehicles in Wilderness Search and Rescue Operations: A Scoping Review,

Vincent-Lambert, C., Pretorius, A. and Van Tonder, B. : “Use of Unmanned Aerial Vehicles in Wilderness Search and Rescue Operations: A Scoping Review,” Wilderness & Environ- mental Medicine, 34(4), 580–588 (2023)

work page 2023
[2]

Unmanned Aerial Vehicles for Search and Rescue: A Survey

Lyu, M., Zhao, Y., Huang, C. and Huang, H. : “Unmanned Aerial Vehicles for Search and Rescue: A Survey”, Remote Sens. 2023, 15, 3266 (2023)

work page 2023
[3]

The Use of UAV’s for Search and Rescue Operations,

Półka, M., Ptak, S. and Kuziora, Ł.: “The Use of UAV’s for Search and Rescue Operations,” Procedia Engineering, 192, 748–752 (2017)

work page 2017
[4]

Automatic Person Detection in Search and Rescue Op- erations Using Deep CNN Detectors,

Sambolek, S. and Ivasic-Kos, M. : “Automatic Person Detection in Search and Rescue Op- erations Using Deep CNN Detectors,” IEEE Access, 9, 37905–37922 (2021)

work page 2021
[5]

Real Time Facial Recognition and Tracking System Using Drones,

Melkumyan, A. and Mkrtchyan, K. : “Real Time Facial Recognition and Tracking System Using Drones,” in 2023 IEEE 20th Consumer Communications & Networking Conference (CCNC), Las Vegas, NV, USA: IEEE, 975–976 (2023)

work page 2023
[6]

Face Tracking for Flying Robot Quadcopter based on Haar Cascade Classifier and PID Controller,

S. Priambodo, A., Arifin, F., Nasuha, A. and Winursito, A. : “Face Tracking for Flying Robot Quadcopter based on Haar Cascade Classifier and PID Controller,” J. Phys.: Conf. Ser., 2111(1), 012046 (2021)

work page 2021
[7]

Visual detection and tracking with UAVs, following a mobile object,

A. Mercado-Ravell, D., Castillo, P. and Lozano, R. : “Visual detection and tracking with UAVs, following a mobile object,” Advanced Robotics, 33(7–8), 388–402 (2019)

work page 2019
[8]

Autonomous UAV Implementa- tion for Facial Recognition and Tracking in GPS-Denied Environments

A. H. Ollachica, D., K. A. Asante, B. and Imamura, H. : “Autonomous UAV Implementa- tion for Facial Recognition and Tracking in GPS-Denied Environments”, IEEE Access, 12, 119464-119487 (2024)

work page 2024

Show all 13 references

[9]

Person Tracking and Frontal Face Capture with UAV,

Shen, Q., Jiang, L. and Xiong, H. : “Person Tracking and Frontal Face Capture with UAV,” in 2018 IEEE 18th International Conference on Communication Technology (ICCT) , Chongqing: IEEE , 1412–1416 (2018)

2018
[10]

Edge Computing-Driven Real-Time Drone Detection Using YOLOv9 and NVIDIA Jetson Nano,

Hakani, R. and Rawat, A. : “Edge Computing-Driven Real-Time Drone Detection Using YOLOv9 and NVIDIA Jetson Nano,” Drones, 8(11), 680 (2024)

2024
[11]

FaceNet: A Unified Embedding for Face Recognition and Clustering,

Schroff, F., Kalenichenko, D. and Philbin, J. : “FaceNet: A Unified Embedding for Face Recognition and Clustering,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 815–823 (2015)

2015
[12]

DJI store: https://store.dji.com/hr/product/tello?vid=38421, last accessed 2024/12/18

2024
[13]

Ultralytics YOLOv11 Documentation: https://docs.ultralytics.com/models/yolo11/, last ac- cessed 2024/12/16

2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Use of Unmanned Aerial Vehicles in Wilderness Search and Rescue Operations: A Scoping Review,

Vincent-Lambert, C., Pretorius, A. and Van Tonder, B. : “Use of Unmanned Aerial Vehicles in Wilderness Search and Rescue Operations: A Scoping Review,” Wilderness & Environ- mental Medicine, 34(4), 580–588 (2023)

work page 2023

[2] [2]

Unmanned Aerial Vehicles for Search and Rescue: A Survey

Lyu, M., Zhao, Y., Huang, C. and Huang, H. : “Unmanned Aerial Vehicles for Search and Rescue: A Survey”, Remote Sens. 2023, 15, 3266 (2023)

work page 2023

[3] [3]

The Use of UAV’s for Search and Rescue Operations,

Półka, M., Ptak, S. and Kuziora, Ł.: “The Use of UAV’s for Search and Rescue Operations,” Procedia Engineering, 192, 748–752 (2017)

work page 2017

[4] [4]

Automatic Person Detection in Search and Rescue Op- erations Using Deep CNN Detectors,

Sambolek, S. and Ivasic-Kos, M. : “Automatic Person Detection in Search and Rescue Op- erations Using Deep CNN Detectors,” IEEE Access, 9, 37905–37922 (2021)

work page 2021

[5] [5]

Real Time Facial Recognition and Tracking System Using Drones,

Melkumyan, A. and Mkrtchyan, K. : “Real Time Facial Recognition and Tracking System Using Drones,” in 2023 IEEE 20th Consumer Communications & Networking Conference (CCNC), Las Vegas, NV, USA: IEEE, 975–976 (2023)

work page 2023

[6] [6]

Face Tracking for Flying Robot Quadcopter based on Haar Cascade Classifier and PID Controller,

S. Priambodo, A., Arifin, F., Nasuha, A. and Winursito, A. : “Face Tracking for Flying Robot Quadcopter based on Haar Cascade Classifier and PID Controller,” J. Phys.: Conf. Ser., 2111(1), 012046 (2021)

work page 2021

[7] [7]

Visual detection and tracking with UAVs, following a mobile object,

A. Mercado-Ravell, D., Castillo, P. and Lozano, R. : “Visual detection and tracking with UAVs, following a mobile object,” Advanced Robotics, 33(7–8), 388–402 (2019)

work page 2019

[8] [8]

Autonomous UAV Implementa- tion for Facial Recognition and Tracking in GPS-Denied Environments

A. H. Ollachica, D., K. A. Asante, B. and Imamura, H. : “Autonomous UAV Implementa- tion for Facial Recognition and Tracking in GPS-Denied Environments”, IEEE Access, 12, 119464-119487 (2024)

work page 2024