A Lightweight Traffic Lights Detection and Recognition Method for Mobile Platform

Abstract

Traffic lights detection and recognition (TLDR) is one of the necessary abilities of multi-type intelligent mobile platforms such as drones. Although previous TLDR methods have strong robustness in their recognition results, the feasibility of deployment of these methods is limited by their large model size and high requirements of computing power. In this paper, a novel lightweight TLDR method is proposed to improve its feasibility to be deployed on mobile platforms. The proposed method is a two-stage approach. In the detection stage, a novel lightweight YOLOv5s model is constructed to locate and extract the region of interest (ROI). In the recognition stage, the HSV color space is employed along with an extended twin support vector machines (TWSVMs) model to achieve the recognition of multi-type traffic lights including the arrow shapes. The dataset, collected in naturalistic driving experiments with an instrument vehicle, is utilized to train, verify, and evaluate the proposed method. The results suggest that compared with the previous YOLOv5s-based TLDR methods, the model size of the proposed lightweight TLDR method is reduced by 73.3%, and the computing power consumption of it is reduced by 79.21%. Meanwhile, the satisfied reasoning speed and recognition robustness are also achieved. The feasibility of the proposed method to be deployed on mobile platforms is verified with the Nvidia Jetson NANO platform.

Early Drought Detection in Maize Using UAV Images and YOLOv8+

“…In the Neck section, three sets of features are produced in total, each set branching into two pathways upon entering the Head: one pathway is dedicated to the computation of detection boxes, with 64 feature channels feeding into the Anchor module to generate these boxes; the other pathway focuses on computing the probabilities of each class, with 80 feature channels passing through the sigmoid activation function to yield the probabilities associated with 80 respective classes. Subsequently, the detection boxes and class probabilities are combined, resulting in a total of 84 feature channels [19]. Three feature maps are generated from bottom to top, with dimensions of 80 × 80 × 80, 40 × 40 × 40, and 20 × 20 × 20.…”

“…The model training parameters were configured as follows: the maximum number of iterations was set to 500, utilizing stochastic gradient descent (SGD) [36] as the optimizer with a momentum value of 0.9. The learning rate adjustment strategy [37] employed cosine annealing decay, as outlined in Equation (19).…”

Coal Mine Rock Burst and Coal and Gas Outburst Perception Alarm Method Based on Visible Light Imagery

“…There are three color parameters in this model: hue (H), saturation (S), and lightness (V). The relational equation for the conversion of color images from the RGB (red, green, blue) color model to the HSV color model is as follows [36] (where R, G, and B are normalized):…”

“…Using the HSV color model in Section 2.1 to analyze whether the average V value (luminance) of the image is less than the set luminance threshold to determine whether the image has abnormal luminance, the relationship [36] is determined as follows:…”

Real-Time Pedestrian Detection with YOLOv7 and Intel MiDaS: A Qualitative Study

“…These anchor boxes were determined through dimension clustering of training data, ensuring accurate representation during model training and ultimately enhancing accuracy [10]. The evolution continued with YOLOv3, detailed in Joseph Redmon and Ali Farhadi's 2018 paper titled "YOLOv3: An Incremental Improvement" [11]. While slightly larger, YOLOv3 maintained commendable speed and accuracy.…”