# Object Detection Survey
###### tags:`Survey`

## Two types of frameworks

---
## Dataset

---
## Evaluation Criteria
* Detection speed `Frames Per Second (FPS)`
* Precision / Recall
---
## Detection Method
### Region Based (Two Stage) Frameworks
* R-CNN
* SPPNet
* Fast R-CNN
* Faster R-CNN
* R-FCN (Region based Fully Convolutional Network)
* proposed using all CONV layers to construct a shared RoI sub-network, and RoI crops are taken from the last layer of CONV features
* FPN
* Mask R-CNN
* first stage : RPN
* second stage : predicting the class and box offset which adds a branch (outputs a binary mask for each RoI).
* Chained Cascade Network & Cascade R-CNN
* multistage classifiers
* first stage : removing large amounts of background
* second stage : classifying the remaining regions.
* Light Head RCNN
---
### Unified (One Stage) Frameworks
* DetectorNet `the first to explore CNNs for object detection`
* OverFeat
* YOLO
* YOLOv2
* SSD
* CornerNet
---
### Overview
---
## Applications
### Pedestrian Detection
#### Difficulties and Challenges
* Small pedestrian
`In Caltech Dataset, 15% of the pedestrians are less than 30 pixels in height`
* Backgrounds are very similar to pedestrians
* Occluded pedestrian
---
### Face Detection
#### Difficulties and Challenges
* Human faces may present a variety of expressions, skin colors, poses, and movements
* Occlusion
* Multi-scale detection
---
### Text Detection
#### Difficulties and Challenges
* Different fonts and languages
* Text rotation and perspective distortion
* Text lines with large aspect ratios and dense layout are difficult to localize
* Broken and blurred characters
---
### Traffic Sign and Traffic Light Detection
#### Difficulties and Challenges
* Detection will be particularly difficult when driving into the sun glare or at night
* Motion blur
* Bad weather
---
### Remote Sensing Target Detection
#### Difficulties and Challenges
* Detection in “big data”
> how to quickly and accurately detect remote sensing targets remains a problem.
* Occluded targets
* Remote sensing images captured by different sensors present a high degree of differences.
---
### A Survey Paper on Object Detection Methods in Image Processing
| Year | Paper | Technique Used | Results |
| ------------------------------ | ------------------------------------------------------------------------------------ | ---------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| 2019 | Fruit Quality Inspection System Using Digital Image Processing Technique | Minimum distance classifier using a statistical distance measure | Pattern recognition using higher order spectral invariants is fast, suited for parallel implementation, and has high immunity to additive Gaussian noise |
| 2018 | Multi-scale deep neural network for salient object detection | Convolution Neural Network ( CNN) | 在多個datasets上，結果表明基於深度學習的模型比其他方法表現更好 |
| <font color="#f00">2018</font> | You Only Look Once: Unified, Real-Time Object Detection | Region-based Convolutional Neural Networks. | The proposed R-CNN based model was able to process images at the rate of 155 frames per second |
| 2018 | Currency Detection and Recognition Based on Deep Learning | 檢測紙幣正面和反面的特徵 | Achieved accuracy rate of 96.6% |
| 2016 | SSD: Single Shot MultiBox Detector | Single deep neural network | Accuracy (76.9%) achieved using Single shot multibox detector |
| 2016 | Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures | Deep CNN based models in the area of medical imaging | Medical imaging cases and was able to arrive at a sensitivity rate of 85% |
| 2015 | A practical implementation of face detection by using Matlab cascade object detector | Viola Jones algorithm | 針對不同的調諧參數，分析了face detectors的性能。 |
| 2014 | Deep Convolutional Neural Networks for Efficient Pose Estimation in Gesture Videos | Convolution Neural Network ( CNN) | The results were found to be encouraging and findings could be utilized by other public datasets |
| 2003 | Face recognition vendor test 2002 : | Face recognition vendor test (FRVT). | 與女性相比，識別男性可能更容易。 |
| 2001 | Rapid object detection using a boosted cascade of simple features | Machine Learning Based Approach|Design is highly effective in processing images at fast pace and achieving outcome in form of higher rate of object detection|

---
Refference
* Deep Learning for Generic Object Detection: A Survey
* Object Detection With Deep Learning: A Review
* Object Detection in 20 Years: A Survey
* A Survey Paper on Object Detection Methods in Image Processing