Embedded Files
The Challenge
The Challenge
The challenge is to perform object detection on the BIRDSAI dataset containing annotated images taken from thermal infrared (TIR) videos. You will be required to create a model that predicts the correct bounding boxes as well as the correct type of object for all the animals or humans present in a sample input image. This will allow for tracking and surveillance for all the objects present in the image. The ground truth bounding boxes will be provided in the dataset thus making this problem a supervised learning problem and your system will be evaluated based on how close your model's predicted bounding boxes are to the ground truth for the test set. The evaluation metric will be the F1 score.
The challenge is to perform object detection on the BIRDSAI dataset containing annotated images taken from thermal infrared (TIR) videos. You will be required to create a model that predicts the correct bounding boxes as well as the correct type of object for all the animals or humans present in a sample input image. This will allow for tracking and surveillance for all the objects present in the image. The ground truth bounding boxes will be provided in the dataset thus making this problem a supervised learning problem and your system will be evaluated based on how close your model's predicted bounding boxes are to the ground truth for the test set. The evaluation metric will be the F1 score.
What's Object Detection ?
What's Object Detection ?
Object Detection involves identifying the presence, location and type of every object in an image. Object detection has applications in many important fields and has been widely used in several problems such as surveillance, autonomous driving, security and face recognition.
Object Detection involves identifying the presence, location and type of every object in an image. Object detection has applications in many important fields and has been widely used in several problems such as surveillance, autonomous driving, security and face recognition.
Object detection requires the prediction of a bounding box as well as it's corresponding label. Therefore, you are required to identify every object in the image and also provide a bounding box denoting it's location in the image. Several evaluation metrics can be used such as mean IOU (Intersection over union) or mAP (mean average precision) to verify how close the predictions are to the ground truth.
Object detection requires the prediction of a bounding box as well as it's corresponding label. Therefore, you are required to identify every object in the image and also provide a bounding box denoting it's location in the image. Several evaluation metrics can be used such as mean IOU (Intersection over union) or mAP (mean average precision) to verify how close the predictions are to the ground truth.
To know more about this, you can read this blog to get started.
To know more about this, you can read this blog to get started.
Role of Object Detection in Surveillance of Protected Areas
Role of Object Detection in Surveillance of Protected Areas
Monitoring of protected areas to curb illegal activities like poaching and animal trafficking is a very crucial conservation task in many regions of South Africa. To augment existing manual patrolling efforts, unmanned aerial surveillance using thermal infrared (TIR) cameras is increasingly being adopted. However, processing large amounts of data from such cameras becomes a challenge especially for monitoring and tracking humans who may be involved in illegal activities.
Monitoring of protected areas to curb illegal activities like poaching and animal trafficking is a very crucial conservation task in many regions of South Africa. To augment existing manual patrolling efforts, unmanned aerial surveillance using thermal infrared (TIR) cameras is increasingly being adopted. However, processing large amounts of data from such cameras becomes a challenge especially for monitoring and tracking humans who may be involved in illegal activities.
Object Detection plays a vital role in this regard as it allows for automatic detection and tracking in TIR images, The outputs of such methods can be used in tracking and detection of people and animals across frames and therefore may be crucial in the prevention of poaching and other illegal activities. There are also several challenging scenarios like scale variations, background clutter due to thermal reflections, large camera rotations, and motion blur which can test the robustness of a object detector. Therefore, a model trained on the dataset must be able to handle such challenging scenarios and make a prediction with precision. These situations make the BIRDSAI dataset an interesting problem to solve with an important real-life application and as participants, you will certainly enjoy this novel challenge.
Object Detection plays a vital role in this regard as it allows for automatic detection and tracking in TIR images, The outputs of such methods can be used in tracking and detection of people and animals across frames and therefore may be crucial in the prevention of poaching and other illegal activities. There are also several challenging scenarios like scale variations, background clutter due to thermal reflections, large camera rotations, and motion blur which can test the robustness of a object detector. Therefore, a model trained on the dataset must be able to handle such challenging scenarios and make a prediction with precision. These situations make the BIRDSAI dataset an interesting problem to solve with an important real-life application and as participants, you will certainly enjoy this novel challenge.
Additional Resources
Additional Resources
Blogs
Object Detection with Deep Learning: The Definitive Guide [Link]
Object Detection and Tracking in 2020 [Link]
Ultimate Guide to Object Detection Using Deep Learning [Link]
A 2019 Guide to Object Detection [Link]
Research Papers
Girshick, Ross. “Fast R-CNN.” ArXiv:1504.08083 [Cs], Sept. 2015. arXiv.org, http://arxiv.org/abs/1504.08083.
Ren, Shaoqing, et al. “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks.” ArXiv:1506.01497 [Cs], Jan. 2016. arXiv.org, http://arxiv.org/abs/1506.01497.
He, Kaiming, et al. “Mask R-CNN.” ArXiv:1703.06870 [Cs], Jan. 2018. arXiv.org, http://arxiv.org/abs/1703.06870.
Liu, Wei, et al. “SSD: Single Shot MultiBox Detector.” ArXiv:1512.02325 [Cs], vol. 9905, 2016, pp. 21–37. arXiv.org, doi:10.1007/978-3-319-46448-0_2, https://arxiv.org/abs/1512.02325
Redmon, Joseph, et al. “You Only Look Once: Unified, Real-Time Object Detection.” ArXiv:1506.02640 [Cs], May 2016. arXiv.org.https://arxiv.org/abs/1506.02640
Page updated
Report abuse