detecting cars in a parking lot using deep learning...single detection layer for cars yolo v3 -...
TRANSCRIPT
![Page 1: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/1.jpg)
Detecting Cars in a Parking Lot using Deep Learning
Samuel OrdoniaMay 2019
![Page 2: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/2.jpg)
Overview
1. Introduction2. Background3. Implementation4. Experiments & Results5. Skynet Demo6. Conclusions7. Q & A
![Page 3: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/3.jpg)
Introduction
● Objective: Detecting cars in a parking lot● Solution: Convolutional Neural Net (CNN) single-shot detector
![Page 4: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/4.jpg)
Technical Challenges
1. Detection of dozens of objects in an image with variable size, color, pose, depth, and occlusion
2. Obtaining thoroughly annotated and accurate training data3. Effective data compilation and validation tool4. Implementation and maintenance of a complex detection model5. Practical application of a parking lot car detector
![Page 5: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/5.jpg)
Car Detection
● Two part problem○ Classification - there is a car in these pixels○ Localization - these pixels are significant
● Detection - these pixels within this image = car
![Page 6: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/6.jpg)
Car Detection: Iconic Example
![Page 7: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/7.jpg)
Other Iconic Detection Examples
![Page 8: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/8.jpg)
Background
![Page 9: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/9.jpg)
CNN Engineering
![Page 10: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/10.jpg)
Building Blocks of the CNN
● Convolution → apply a weighted kernel (i.e., filter) across an input tensor to derive feature maps.
● Convolutional Layer○ Convolution transformation○ Non-linear activation function○ Pooling Layer
● Kernel● Stride● Padding
![Page 11: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/11.jpg)
Convolution Layer Breakdown
![Page 12: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/12.jpg)
Intersection over Union (IOU)
● Objectness score.● Comparison of the predicted box to the ground truth one.
IOU = Area of Intersection / Total Area
![Page 13: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/13.jpg)
Poor IOU
![Page 14: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/14.jpg)
Excellent IOU
![Page 15: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/15.jpg)
Anchor Boxes
● Simplify bounding box regression.● Compute IOU based on predetermined anchor boxes instead of the ground
truth boxes.● Predict multiple classes at the same centroid pixel.
![Page 16: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/16.jpg)
Anchor Boxes Derivation
● Derived from k-means clustering of ground truth bounding boxes.● Determines centroids based on the labeled dataset’s height and width
parameters.● No association of (x, y) positions in the image are accounted for.
![Page 17: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/17.jpg)
Anchor Boxes Example
![Page 18: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/18.jpg)
Anchor Boxes Example
![Page 19: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/19.jpg)
Non-max Suppression
● Bounding box regression approach.● Maximize intersection of predicted over ground truth bounding boxes.● Discard excess bounding boxes of high similarity (IOU).
![Page 20: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/20.jpg)
Non-max Suppression
![Page 21: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/21.jpg)
You Only Look Once (YOLO)
● Fully convolutional net → faster and less memory heavy than a fully connected net.
● Performs classification and localization in the same step → faster than multi-stage deep learning models.
![Page 22: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/22.jpg)
YOLO v2 - Faster, Better
● Switched to anchor boxes and NMS for detection regression.● Optimal runtime with Darknet-19 feature extraction CNN.● Single detection layer for cars
![Page 23: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/23.jpg)
YOLO v3 - Three Layers of Detection
● Much deeper feature extraction → higher accuracy, but slower runtime.● Three detection layers routing different feature maps from the Darknet-53
feature extraction portion. ● Detects cars at different depths.
![Page 24: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/24.jpg)
YOLO v3
![Page 25: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/25.jpg)
Implementation
![Page 26: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/26.jpg)
Vision
● Open CV Docker application for parking lot data compilation, processing, and validation.
● Compilation → uniform format for training the CNN models● Validation → verify accuracy of the ground truth labeling
![Page 27: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/27.jpg)
Vision: Open CV
● Computer Vision framework with Python bindings.● Highly optimized in C.● Installed binaries from the framework → wreak havoc with the host machine’s
potentially existing versions of those executables or libraries.
![Page 28: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/28.jpg)
Vision: Docker
● Linux containerization as a service.● Safely install and execute Open CV within the Docker application. Host
machine’s existing programs are untouched. ● Concise syntax and well documented features.
![Page 29: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/29.jpg)
Vision Architecture
![Page 30: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/30.jpg)
Vision: Bounding Box Annotation
![Page 31: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/31.jpg)
Vision: Bounding Box Annotation
![Page 32: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/32.jpg)
Research Environment
● Pytorch deep learning application powered by NVIDIA’s CUDA GPU integration.
● Models defined in configuration (cfg) files.● Factory-like Pytorch logic generates a model and its layers.
![Page 33: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/33.jpg)
Research Environment Architecture
![Page 34: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/34.jpg)
Research Environment: Pytorch
● Pytorch is a Python deep learning framework for research and production. ● Flexible & concise syntax abstracts away enough lower-level logic involving
neural nets.● Imperative execution → easily debug as deep learning scripts are executed.● Seamless & straightforward integration with NVIDIA CUDA framework.
![Page 35: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/35.jpg)
Research Environment: Model Configurations
● Standardized file format (cfg) to represent model definitions.● Blocks contain hyperparameters or layer-specific parameters.● Used throughout the deep learning community → independent of language
or framework
![Page 36: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/36.jpg)
Skynet: Practical Application
● Docker application to execute statistical analysis on detection results.● Invokes the Research Environment’s Pytorch Detector, though it is decoupled
from the core training environment.● Executable on a device with or without a GPU.
![Page 37: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/37.jpg)
Skynet: Car Count Time Series
● Car detection across several images over time.● Plots car counts in the parking lot over time.
![Page 38: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/38.jpg)
Experiments & Results
![Page 39: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/39.jpg)
Training Data
● CARPK○ Top-down drone captured parking lot images○ Cars have different poses but same sizes○ Minimal occlusion○ Fully annotated ground truth bounding boxes
● CNR○ Fixed camera angle covering the parking lot○ Cars have different poses and depth○ More occlusions behind trees and other cars○ Not all cars are labeled with bounding boxes
![Page 40: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/40.jpg)
Input Image Example
![Page 41: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/41.jpg)
Input Image Example
![Page 42: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/42.jpg)
Models
● YOLO v2 (Darknet-19 + one detection layer)● YOLO v3 (Darknet-53 + three detection layers)● Modified YOLO v3 (Darknet-53 + four detection layers)
![Page 43: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/43.jpg)
YOLO v2: One Detection Layer
● Single detection layer with anchor boxes.● Fastest of the CNN models in this project due to having the fewest layers.● Least accurate because of single detection layer.
![Page 44: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/44.jpg)
CNN Detection Result: YOLO v2
![Page 45: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/45.jpg)
CNN Detection Result: YOLO v2
![Page 46: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/46.jpg)
YOLO v3: Triple the Detection Layers
● Designed to detect a range of differently sized objects.● Three detection layers at different levels of upsample.● Routing layers concatenate a feature map with the “current” tensor.
![Page 47: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/47.jpg)
CNN Detection Result: YOLO v3
![Page 48: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/48.jpg)
CNN Detection Result: YOLO v3
![Page 49: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/49.jpg)
Modified YOLO v3: Fourth Detection Layer
● Same Darknet-53 feature extraction portion.● Routed a fourth feature map to a corresponding fourth detection layer. ● Computed additional anchor boxes.
![Page 50: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/50.jpg)
CNN Detection Result: 4th Detection Layer
![Page 51: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/51.jpg)
CNN Detection Result: 4th Detection Layer
![Page 52: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/52.jpg)
CNN Detection Results Evolution
![Page 53: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/53.jpg)
YOLO v2 YOLO v3 YOLO the 4th
![Page 54: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/54.jpg)
YOLO v2 YOLO v3
YOLO the 4th
![Page 55: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/55.jpg)
Skynet Demo
https://www.youtube.com/watch?v=3brf_r1hnr0&t=38s
![Page 56: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/56.jpg)
Conclusions
● Final results with the fourth detection layer modified YOLO CNN● Research Environment: Overall parking lot car detection
○ Top-down detection is near-perfect ○ Fixed-camera detection progressively improves with more detection layers
● Skynet: Parking lot capacity monitoring○ Top-down drone coverage is most ideal in detection accuracy○ Fixed-angle is effective at parking lot capacity monitoring
![Page 57: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/57.jpg)
Next Steps
● More comprehensively labeled parking lot datasets● Beyond Darknet-53 → Deeper feature extraction portion● Skynet video stream support● Skynet web application● Beyond parkings lots → inventory management
![Page 58: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/58.jpg)
Thank You
![Page 59: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/59.jpg)
Appendix
![Page 60: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/60.jpg)
Appendix A - References
1. J. Redmon, et al, “You Only Look Once: Unified, Real-Time Object Detection,” arXiv, May 2016. [Online]. Available: https://arxiv.org/pdf/1506.02640.pdf
2.
![Page 61: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/61.jpg)
Appendix B - Convolution Detailed Overview
● Vertical edge detection example● Detailed look at padding and stride● Convolution vs cross-correlation
![Page 62: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/62.jpg)
Appendix B - Vertical Edge Detection Example
![Page 63: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/63.jpg)
Appendix B - Padding & Stride
![Page 64: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/64.jpg)
Appendix B - Convolution vs Cross-Correlation
● Convolution in mathematics literature refers to a slightly different operation than Convolution done in deep learning
● Convolution in deep learning ~ Cross-correlation in mathematics
![Page 65: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/65.jpg)
Appendix B: Pooling
![Page 66: Detecting Cars in a Parking Lot using Deep Learning...Single detection layer for cars YOLO v3 - Three Layers of Detection Much deeper feature extraction → higher accuracy, but slower](https://reader033.vdocument.in/reader033/viewer/2022050421/5f909f8389c97744cf2d2bc0/html5/thumbnails/66.jpg)
Appendix C - Hardware Solution Shortcomings