Smartphone‐based object recognition with embedded machine learning intelligence for unmanned aerial vehicles

et al. 2021

Neural Comput & Applic

Self Cite

This paper proposes an acceleration technique to minimise the unnecessary operations on a state-of-the-art machine learning model and thus to improve the processing speed while maintaining the accuracy. After the study of the main bottlenecks that negatively affect the performance of convolutional neural networks, this paper designs and implements a discarding technique for YOLOv3-based algorithms to increase the speed and maintain accuracy. After applying the discarding technique, YOLOv3 can achieve a 22% of improvement in terms of speed. Moreover, the results of this new discarding technique were tested on Tiny-YOLOv3 with three output layers on an autonomous vehicle for pedestrian detection and it achieved an improvement of 48.7% in speed. The dynamic discarding technique just needs one training process to create the model and thus execute the approach, which preserves accuracy. The improved detector based on the discarding technique is able to readily alert the operator of the autonomous vehicle to take the emergency brake of the vehicle in order to avoid collision and consequently save lives.

Section: Implementation Of the Approach To A Real Use Casementioning

confidence: 99%

A dynamic discarding technique to increase speed and preserve accuracy for YOLOv3

et al. 2021

Neural Comput & Applic

Self Cite

“…Others deploy OpenCV, which combines machine learning with computer vision. These studies, however, did not provide all the necessary metrics including model size to make the comparisons [21].…”

Section: Analysis Of Previous Smartphone Object Recognition Workmentioning

confidence: 99%

“…As a result, Snapdragon was not evaluated in this section. The authors of [21] provided two reasons in favour of TFL when it comes to tracking. First of all, TensorFlow is faster than OpenCV when tracking an object.…”

Section: Trackermentioning

confidence: 99%

“…For each step, the selection of the best mobile platform based on evaluation is required to obtain the best system performance. A platform is typically evaluated based on criteria including accuracy, speed, RAM usage and model size [21]. Image recognition is a computational expensive process able to mark with a bounding box the objective in an image; however, although object tracking is much less computational expensive, it needs the initial coordinates of the object to follow its movements in the next frames.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Smartphone-based real-time object recognition architecture for portable and constrained systems

et al. 2021

J Real-Time Image Proc

Self Cite

Machine learning algorithms based on convolutional neural networks (CNNs) have recently been explored in a myriad of object detection applications. Nonetheless, many devices with limited computation resources and strict power consumption constraints are not suitable to run such algorithms designed for high-performance computers. Hence, a novel smartphone-based architecture intended for portable and constrained systems is designed and implemented to run CNN-based object recognition in real time and with high efficiency. The system is designed and optimised by leveraging the integration of the best of its kind from the state-of-the-art machine learning platforms including OpenCV, TensorFlow Lite, and Qualcomm Snapdragon informed by empirical testing and evaluation of each candidate framework in a comparable scenario with a high demanding neural network. The final system has been prototyped combining the strengths from these frameworks and led to a new machine learning-based object recognition execution environment embedded in a smartphone with advantageous performance compared with the previous frameworks.

“…In summary, none of the above studies has covered or considered the essential factors of human detection [41,42]. Moreover, none of the previous studies has considered the best trade-off between speed and accuracy suitable for our top-view human detection use case.…”

Section: Previous Workmentioning

confidence: 99%

Machine-learning-based top-view safety monitoring of ground workforce on complex industrial sites

et al. 2021

Neural Comput & Applic

Self Cite

Telescopic cranes are powerful lifting facilities employed in construction, transportation, manufacturing and other industries. Since the ground workforce cannot be aware of their surrounding environment during the current crane operations in busy and complex sites, accidents and even fatalities are not avoidable. Hence, deploying an automatic and accurate top-view human detection solution would make significant improvements to the health and safety of the workforce on such industrial operational sites. The proposed method (CraneNet) is a new machine learning empowered solution to increase the visibility of a crane operator in complex industrial operational environments while addressing the challenges of human detection from top-view on a resource-constrained small-form PC to meet the space constraint in the operator’s cabin. CraneNet consists of 4 modified ResBlock-D modules to fulfill the real-time requirements. To increase the accuracy of small humans at high altitudes which is crucial for this use-case, a PAN (Path Aggregation Network) was designed and added to the architecture. This enhances the structure of CraneNet by adding a bottom-up path to spread the low-level information. Furthermore, three output layers were employed in CraneNet to further improve the accuracy of small objects. Spatial Pyramid Pooling (SPP) was integrated at the end of the backbone stage which increases the receptive field of the backbone, thereby increasing the accuracy. The CraneNet has achieved 92.59% of accuracy at 19 FPS on a portable device. The proposed machine learning model has been trained with the Standford Drone Dataset and Visdrone 2019 to further show the efficacy of the smart crane approach. Consequently, the proposed system is able to detect people in complex industrial operational areas from a distance up to 50 meters between the camera and the person. This system is also applicable to the detection of any other objects from an overhead camera.