WEEK 7: TRY DIFFERENT OBJECT DETECTION METHOD & OBJECT TRACKING

Date: 17 - 21 April 2023

Content:
During week 7, my focus shifted to learning object tracking and finding alternative methods for object detection. I started by exploring the concept of object tracking and its applications in real-time visual analysis. Optical flow algorithms, which to analyze pixel motion between frames and meanshift algorithms, which updates object locations based on pixel distributions, were among the techniques I studied.

Additionally, I encountered a challenge when using Haarcascade methor for detecting Red Tilapia Fish. To address this limitation, I discovered a Python library called 'Detecto' which facilitated the development of a custom-trained object detection model. I am using annotation tool called 'MAKESENSE.AI' to annotate the fish in images and trained the model using the Faster R-CNN ResNet-50 FPN architecture known for effectiveness in object detection. After several rounds of testing, fine-tuning, and iteration, I achieved remarkable results in detecting Red Tilapia Fish in image.

Conclusion:
Overall, week 7 involved an exploration of object tracking techniques and an adaptation to alternative methods for object detection. By using 'Detecto', I successfully developed a custom-trained object detection model for Red Tilapia Fish. For the next week, I want to explore more about 'Detecto' and use it to try on Real-Time Video.

Appendix:
Figure 1 shows the output fish detection by using Haarcascade Method, as shown in the Figure 1, even after doing some thresholding, it still cannot detecting the whole fish.

Figure 1: Haarcascade Method Output

Figure 2 shows the output of fish detection by using 'Detecto' and Fast R-CNN Method

Figure 2: Detecto Output




Comments

Popular posts from this blog

WEEK 3: IMAGE BASICS AND OPENCV

WEEK 6: LEARNING OBJECT DETECTION FOR MY SYSTEM

WEEK 2: FYP WORKSHOP & COMPUTER VISION INTRODUCTION