Real-time vehicle overload detection method based on convolutional neural network
The present disclosure provides a real-time vehicle overload detection method based on a convolutional neural network (CNN). The present disclosure detects a road driving vehicle in real time with a CNN method and a you only look once (YOLO)-V3 detection algorithm, detects the number of wheels to obtain the number of axles, detects a relative wheelbase, compares the number of axles and the relative wheelbase with a national vehicle load standard to obtain a maximum load of the vehicle, and compares the maximum load with an actual load measured by a piezoelectric sensor under the vehicle, thereby implementing real-time vehicle overload detection. The present disclosure has desirable real-time detection, can implement no-parking vehicle overload detection on the road, and avoids potential traffic congestions and road traffic accidents.
1. A real-time vehicle overload detection method based on a convolutional neural network (CNN), comprising:
step 1: preparing a visual object classes (VOC) dataset, comprising:
acquiring a plurality of images of each of vehicles with different axle numbers and wheelbases, and labeling each wheel and a vehicle body in each of the plurality of images using a labeling tool as the VOC dataset;
step 2: configuring a training environment for a YOLO-V3 object detection network model, comprising:
constructing the YOLO-V3 object detection network model with a darknet deep learning framework in a ubuntu system;
step 3: training the YOLO-V3 object detection network model, comprising:
training the YOLO-V3 object detection network model based on the VOC dataset using a YOLO-V3 object detection algorithm, and simplifying the YOLO-V3 object detection network model using a pruning algorithm; and
step 4: uploading the trained YOLO-V3 object detection network model to a server, acquiring an image of a vehicle in real time by a camera and uploading the image to the server, detecting a number of axles and a relative wheelbase of the vehicle using the trained YOLO-V3 object detection network model, and obtaining a theoretical maximum load of the vehicle by comparing a detection result with a national vehicle load standard GB1589-2016; obtaining a true load of the vehicle through a piezoelectric sensor under a road; and determining whether the vehicle is overloaded by comparing the theoretical maximum load with the true load.
2. The real-time vehicle overload detection method according to claim 1 , wherein step 3 comprises:
step 3.1: pre-training the YOLO-V3 object detection network model with Darknet53, and training the YOLO-V3 object detection network model using the VOC dataset prepared in step 1, wherein the Darknet53 is composed of a series of 1×1 and 3×3 convolutional layers, with a total of 53 layers, and each convolutional layer is followed by a batch normalization (BN) layer and LeakyReLU layer; and
step 3.2: sparsely training the YOLO-V3 object detection network model, performing channel pruning on the CNN according to a proportion or a set threshold, and performing iterative pruning according to a precision of a pruned network until a detection precision meets a requirement.
3. The real-time vehicle overload detection method according to claim 1 , wherein detecting the number of axles of the vehicle comprises, by using coordinate information of a wheel bounding box and a vehicle body bounding box, calculating only a number of wheels in the vehicle body bounding box.
4. The real-time vehicle overload detection method according to claim 2 , wherein performing channel pruning on the network CNN according to the proportion or the set threshold, comprises:
sorting all channels based on characteristics of numerical values comprising an average of the intrinsic parameters, an L1 norm and an L2 norm;
pruning, according to a sorted result and the proportion or the set threshold, channels less affecting the number of axles of the vehicle and a detection precision of the wheel; and
using γ parameters of the BN layers as sparse factors, performing L1 regularization on the γ parameters such that a part of γ parameters approach to 0, and pruning convolution kernels having a γ parameter less than the set threshold.
5. The real-time vehicle overload detection method according to claim 1 , wherein detection effect of the trained YOLO-V3 object detection network model is evaluated based on object detection evaluation indexes comprising a precision and a recall, with equations:
Precision
=
T
P
T
P
+
F
P
,
Recall
=
T
P
T
P
+
F
N
wherein, TP represents a true positive indicating that a detection result of an object is a wheel and the object is the wheel; FP represents a false positive indicating that a detection result of the object is the wheel, but the object is not the wheel; and FN represents a false negative, indicating that a detection result of the object is not the wheel, but the object is the wheel.
6. The real-time vehicle overload detection method according to claim 5 , wherein network performance of the trained YOLO-V3 object detection network model is evaluated based on an average precision (AP), wherein the AP is calculated as follows:
AP=∫ 0 1 P ( r ) dr
wherein, P represents the precision, r represents the recall, and P is a function with r as a parameter; a result obtained is an area enclosed by a Precision-Recall curve; and a higher AP value indicates a better performance of the trained YOLO-V3 object detection network model for detecting the number of axles and the wheels of the vehicle.