Figure 1.
Illustration of the concepts of detection and classification.
Figure 1.
Illustration of the concepts of detection and classification.
Figure 2.
Illustration of the architecture of a CNN.
Figure 2.
Illustration of the architecture of a CNN.
Figure 3.
Comparison of threshold methods.
Figure 3.
Comparison of threshold methods.
Figure 4.
Use of Adaptive Threshold to determine degradation status: (a) Outcome for a crosswalk exhibiting signs of wear; (b) Outcome for a crosswalk without signs of wear; (c) Outcome for a crosswalk exhibiting a considerable degree of wear.
Figure 4.
Use of Adaptive Threshold to determine degradation status: (a) Outcome for a crosswalk exhibiting signs of wear; (b) Outcome for a crosswalk without signs of wear; (c) Outcome for a crosswalk exhibiting a considerable degree of wear.
Figure 5.
Example of a question asked on the form.
Figure 5.
Example of a question asked on the form.
Figure 6.
Example of the results obtained in the questionnaire.
Figure 6.
Example of the results obtained in the questionnaire.
Figure 7.
Example of how the increase in pixels can influence the results.
Figure 7.
Example of how the increase in pixels can influence the results.
Figure 8.
Illustration from CNN CSPDarknet53-tiny. Adapted from [
9].
Figure 8.
Illustration from CNN CSPDarknet53-tiny. Adapted from [
9].
Figure 9.
Architecture of the YOLOv4-tiny model. Adapted from [
11].
Figure 9.
Architecture of the YOLOv4-tiny model. Adapted from [
11].
Figure 10.
Example of a captured image.
Figure 10.
Example of a captured image.
Figure 11.
Illustration of objects captured by a 90° camera vs. a 160° camera.
Figure 11.
Illustration of objects captured by a 90° camera vs. a 160° camera.
Figure 12.
Diagram of how the first approach works.
Figure 12.
Diagram of how the first approach works.
Figure 13.
Examples of Adaptive Threshold detection: (a) Bounding boxes of three crosswalks (1)–(3) detected by the YOLOv4-tiny model; (b) Result of using the Adaptive Threshold method to determine the degradation state of the three crosswalks (1)–(3).
Figure 13.
Examples of Adaptive Threshold detection: (a) Bounding boxes of three crosswalks (1)–(3) detected by the YOLOv4-tiny model; (b) Result of using the Adaptive Threshold method to determine the degradation state of the three crosswalks (1)–(3).
Figure 14.
Examples of Adaptive Threshold detection in scenes without shadows: (a) Example 1: detection (1), identification (2), and classification (3) process; (b) Example 2: detection (1), identification (2), and classification (3) process.
Figure 14.
Examples of Adaptive Threshold detection in scenes without shadows: (a) Example 1: detection (1), identification (2), and classification (3) process; (b) Example 2: detection (1), identification (2), and classification (3) process.
Figure 15.
Comparison between the two approaches in two different examples: Example (a): (1) and (2) shows the detection using the first approach considered—YOLOv4-tiny and Adaptive Threshold. (3) shows the result obtained under the same conditions but using the second approach—YOLOv4-tiny with three classes. Example (b): (1) and (2) shows the detection using the first approach considered—YOLOv4-tiny and Adaptive Threshold. (3) shows the result obtained under the same conditions but using the second approach—YOLOv4-tiny with three classes.
Figure 15.
Comparison between the two approaches in two different examples: Example (a): (1) and (2) shows the detection using the first approach considered—YOLOv4-tiny and Adaptive Threshold. (3) shows the result obtained under the same conditions but using the second approach—YOLOv4-tiny with three classes. Example (b): (1) and (2) shows the detection using the first approach considered—YOLOv4-tiny and Adaptive Threshold. (3) shows the result obtained under the same conditions but using the second approach—YOLOv4-tiny with three classes.
Figure 16.
The architecture of the prototype was developed.
Figure 16.
The architecture of the prototype was developed.
Figure 17.
Schematic of the electrical circuit of the various hardware components.
Figure 17.
Schematic of the electrical circuit of the various hardware components.
Figure 18.
Hardware components of the prototype developed.
Figure 18.
Hardware components of the prototype developed.
Figure 19.
Use case diagram.
Figure 19.
Use case diagram.
Figure 20.
Flowchart of how the Raspberry Pi works.
Figure 20.
Flowchart of how the Raspberry Pi works.
Figure 21.
Structure required in documents.
Figure 21.
Structure required in documents.
Figure 22.
Endpoints on the API.
Figure 22.
Endpoints on the API.
Figure 23.
Example of a response from the Geoapify API.
Figure 23.
Example of a response from the Geoapify API.
Figure 24.
Sequence diagram between API and Raspberry PI.
Figure 24.
Sequence diagram between API and Raspberry PI.
Figure 25.
Illustration of the mechanism for identifying the distance between coordinates.
Figure 25.
Illustration of the mechanism for identifying the distance between coordinates.
Figure 26.
Sequence diagram of communication between the API and the CrosSafe application.
Figure 26.
Sequence diagram of communication between the API and the CrosSafe application.
Figure 27.
Mockup Home page.
Figure 27.
Mockup Home page.
Figure 28.
Mockup Dashboard page.
Figure 28.
Mockup Dashboard page.
Figure 29.
Crosswalk counting component by state of degradation.
Figure 29.
Crosswalk counting component by state of degradation.
Figure 30.
Interactive map in the application.
Figure 30.
Interactive map in the application.
Figure 31.
Table with additional information on crosswalks detected using the pagination method.
Figure 31.
Table with additional information on crosswalks detected using the pagination method.
Figure 32.
Popup for confirmation of crosswalk repair.
Figure 32.
Popup for confirmation of crosswalk repair.
Figure 33.
Example of a success message following confirmation of a crosswalk repair.
Figure 33.
Example of a success message following confirmation of a crosswalk repair.
Figure 34.
An example of a crosswalk classified as severely worn.
Figure 34.
An example of a crosswalk classified as severely worn.
Figure 35.
Example of the number of documents received by Firestore.
Figure 35.
Example of the number of documents received by Firestore.
Figure 36.
Illustration of the route taken.
Figure 36.
Illustration of the route taken.
Figure 37.
Detection and classification time required.
Figure 37.
Detection and classification time required.
Figure 38.
An image containing a crosswalk slightly diagonally.
Figure 38.
An image containing a crosswalk slightly diagonally.
Figure 39.
Visualization of the data transmitted by the Raspberry PI in the CrosSafe web application.
Figure 39.
Visualization of the data transmitted by the Raspberry PI in the CrosSafe web application.
Table 1.
Wear interval calculated using Adaptive Threshold.
Table 1.
Wear interval calculated using Adaptive Threshold.
Intervals | State of Degradation |
---|
<30 | No wear |
[30, 50] | Moderate wear |
>50 | Severe wear |
Table 2.
Total number of labels per class.
Table 2.
Total number of labels per class.
Class | State of Degradation |
---|
No wear | 555 |
Moderate wear | 413 |
Severe wear | 471 |
Table 3.
Comparison of the results obtained between the model presented in [
8] and the one trained for this project (*).
Table 3.
Comparison of the results obtained between the model presented in [
8] and the one trained for this project (*).
Model | Images | Input Size | mAP (%) |
---|
YOLOv4-tiny [8] | 642 | 608 × 608 | 87 |
YOLOv4-tiny * | 1182 | 608 × 608 | 90 |
Table 4.
Results of the YOLOv4-tiny model for the 3-class dataset.
Table 4.
Results of the YOLOv4-tiny model for the 3-class dataset.
Class | mAP (%) | Overall (%) |
---|
No wear | 82.13 | 71.21 |
Moderate wear | 54.72 |
Severe wear | 76.96 |
Table 5.
Raspberry Pi consumption.
Table 5.
Raspberry Pi consumption.
Hardware | Watts | Volts | Amps |
---|
Raspberry PI 5 8 GB with cooler | 27 | 5 | 6 |
Raspberry PI 5 8 GB without cooler | 27 | 5 | 5 |
Table 6.
Comparison of real classifications with those made by the YOLOv4-tiny model.
Table 6.
Comparison of real classifications with those made by the YOLOv4-tiny model.
Class | Real Classification | YOLOv4-Tiny |
---|
No wear | 9 | 9 |
Moderate wear | 3 | 2 |
Severe wear | 5 | 4 |
Total | 17 | 15 |