Table 1:
Literature review
| Reference | Method/algorithm used | Merits | Demerits |
|---|---|---|---|
| [1] | YOLO and OpenCV | Provides real-time object detection and visual replacement for the blind. | YOLO may struggle with small or distant objects, and OpenCV's accuracy can vary based on lighting conditions and object complexity. |
| [2] | TensorFlow API, CNN, SSD, and MobileNet V2 | Achieves high accuracy without needing an Internet connection. | May require significant computational resources, especially for training the model. |
| [3] | CNN | Utilizes data augmentation to achieve a 94% accuracy rate for banknote recognition. | Edge-detected images negatively affect accuracy, indicating a need for larger datasets and varied lighting conditions. |
| [4] | CNN | Improves runtime and accuracy for heart rate estimation in large groups. | Specific details about the adapted algorithm and its implementation are needed for a thorough evaluation. |
| [5] | YOLO and SSD | Uses Raspberry Pi devices for a compact travel aid, demonstrating real-world implementation. | The system may face limitations in detecting objects in complex environments or under varying lighting conditions. |
| [6] | CNN and LSTM | Uses Braille and sound bite hearing devices for communication, achieving high accuracy. | The system's effectiveness may depend on the user's familiarity and comfort with Braille. |
| [9] | FSP algorithm | Utilizes a panoramic camera for pedestrian trajectory prediction, improving real-time performance. | The algorithm's accuracy and performance in dynamic or crowded environments need further evaluation. |
| [10] | CNN | Achieves 98% accuracy in diagnosing glaucoma using retinal images. | The effectiveness of the technique in clinical settings and its generalizability to diverse populations need further validation. |
| [11] | Four-layered CNN | Detects and classifies objects with high accuracy and low response time. | The device's performance may vary based on the complexity of the environment and the types of objects present. |
| [12] | CNN and fuzzy logic | Provides auditory feedback for obstacle detection, enhancing interaction with surroundings. | The computational complexity of the algorithms may affect real-time performance. |
| [13] | CNN | Uses smart glasses to identify medicine, showing promise for real-world implementation. | The system's accuracy and reliability in identifying specific medications need further validation. |
Figure 1:
System block diagram.
Figure 2:
Flowchart of system/CNN layers. CNN, convolutional neural network.
Figure 3:
Dataset of Dolo medicine.
Figure 4:
Dataset of Volini medicine.
Figure 5:
Dataset of Strepsils medicine.
Figure 6:
Epochs result for the inception model.
Figure 7:
CNN architecture. CNN, convolutional neural network.
Figure 8:
Summary of parameters and trained CNN model. CNN, convolutional neural network.
Figure 9:
Output results for Strepsils.
Figure 10:
Output results for Volini gel.
Figure 11:
Output results for Dolo.
Figure 12:
Output results for Jovees shampoo.
Table 2:
Success rate of the proposed system
| Succeed | Failed | |||
|---|---|---|---|---|
| 1. | Strepsils | 46 | 4 | 92 |
| 2. | Volini gel | 48 | 2 | 96 |
| 3. | Dolo | 47 | 3 | 94 |
| Success rate≥ | 94 |
Table 3:
Comparative study for the success rate of the proposed system with existing work
| Reference | Method used | Dataset size | Accuracy (%) | Medicines handled | Key advantage |
|---|---|---|---|---|---|
| [13] | CNN | ∼2000 images | 92 | 3 | Smart glasses integration |
| [6] | CNN + LSTM | ∼3000 images | 94 | 5 | Includes Braille output |
| [3] | CNN | Augmented banknote data | 94 | Not medicine-specific | Focused on money recognition |
| Proposed system | Inception CNN | 16,000 images | 96.98 | 4 | Audio output + medication reminder |