|
| Paper | Objective and approach | Algorithm | Dataset | Accuracy | Remarks |
|
| Chiverton [30] | Background subtraction is used to identify bikes, and SVM is used to identify bikers without a helmet. | Support vector machine | Self-generated | N/A | High computation cost because the system determines the full frame for the detection of helmet. |
| Silva et al. [31] | A multilayer perception model is used to differentiate between different objects. The hybrid descriptor is based on the local binary operator to extract features. | Hough transforms with SVM | Self-generated | 94.23% | Low-quality images are used. Descriptors give multiple attributes, which makes classification difficult. |
| Silva et al. [32] | Hough transformation and histogram oriented gradient help extract the image’s attributes. | Hough transformation and histogram oriented gradient | Self-generated | 91.37% | N/A |
| Waranusast et al. [33] | K-nearest neighbor classifier is applied to identify bikers not wearing a helmet. | Machine vision | Self-generated | 74% | Accuracy is low. |
| Dahiya et al. [34] | Classification is done by using a binary classifier and visual features. | HOG, SIFT, LBP, and SVM | Self-generated | 93.80 | The system is not trained to classify bikers wearing the scarf instead of the helmet. |
| Boonsirisumpun et al. [35] | The single shot multibox detector (SSD) technique for detecting motorcyclists not wearing helmets is used. | Google Net, MobileNet, VGG19, and VGG16 | Self-generated | 85.19% | The system is not trained to classify bikers wearing the scarf instead of the helmet. |
| Raj et al. [36] | The researchers used HOG to identify bikers and CNN to detect helmet violators and number plate. | CNN | Self-generated | 94.70% | N/A |
| Wu et al. [37] | YOLOv3 and YOLO-dense backbone are used to detect bikers without a helmet. | Deep learning | Dataset collected from two sources, i.e. self-generated and internet | YOLOv3: 95.15% and YOLO-dense backbone: 97.59% | For YOLOv3, mean average precision is 95.15%, and for the YOLO-dense backbone, it is 97.59% |
| Siebert and Lin [38] | A deep learning algorithm is used to detect bikers without a helmet. | ResNet50 | Self-generated | 72.8% | Accuracy is low, which can be improved. |
| Vishnu et al. [39] | CNN is used for helmet detection. | CNN | Self-generated | Accuracy is 93% | N/A |
| Mistry et al. [40] | YOLOv2 is used to detect bikers without a helmet. | Deep learning | COCO dataset | Accuracy is 94.7% | N/A |
| Afzal et al. [41] | Faster R-CNN is used to detect bikers without a helmet. | Faster R-CNN | Self-generated dataset | Accuracy is 97.26% | N/A |
|
