CN114639084A - Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm - Google Patents

Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm Download PDF

Info

Publication number
CN114639084A
CN114639084A CN202210182687.XA CN202210182687A CN114639084A CN 114639084 A CN114639084 A CN 114639084A CN 202210182687 A CN202210182687 A CN 202210182687A CN 114639084 A CN114639084 A CN 114639084A
Authority
CN
China
Prior art keywords
ssd
algorithm
vehicle
loss function
picture
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210182687.XA
Other languages
Chinese (zh)
Inventor
王祥雪
洪曙光
林焕凯
刘双广
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Gosuncn Technology Group Co Ltd
Original Assignee
Gosuncn Technology Group Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Gosuncn Technology Group Co Ltd filed Critical Gosuncn Technology Group Co Ltd
Priority to CN202210182687.XA priority Critical patent/CN114639084A/en
Publication of CN114639084A publication Critical patent/CN114639084A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Abstract

The invention belongs to vehicle detection of road cooperative road side ends, and particularly relates to a road side end vehicle sensing method based on an SSD (solid State disk) improved algorithm. The improved feature extraction module based on the cross-stage partial connection mode and the residual error connection fusion of the resnet18 has stronger feature expression capability and reduces the time consumption of forward calculation. The SSD detection algorithm based on the caffe framework improves the positioning loss function, fuses the Complete-IOU positioning loss function, improves the accuracy of the detection frame, and enables the model to be converged more quickly.

Description

Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm
Technical Field
The invention belongs to vehicle detection of road cooperative road side ends, and particularly relates to a road side end vehicle sensing method based on an SSD (solid State disk) improved algorithm.
Background
The vehicle-road cooperation is a safe, efficient and environment-friendly road traffic system which adopts the advanced wireless communication, new generation internet and other technologies, implements vehicle-road dynamic real-time information interaction in all directions, develops vehicle active safety control and road cooperative management on the basis of full-time dynamic traffic information acquisition and fusion, fully realizes effective cooperation of human and vehicle roads, ensures traffic safety and improves traffic efficiency. The technical key points to be urgently needed to be broken through at present are road perception, identification of a road and a road environment, identification of a position, a speed first-level motion direction of a road participating main body, identification of an abnormal event occurring on the road and further digital road foundation for an automatic driving vehicle. The video perception technology based on camera imaging can reflect the real scene of the road most truly, and the accuracy of the road (including road surface conditions, mark lines, trees, building facilities, spilled objects, road surface collapse and the like), road participation bodies (people, vehicles, road environments and road abnormal events) can be identified by adopting a computer vision algorithm.
In the conventional image target detection algorithm period, a method based on artificial design feature extraction is mainly used, a sliding window is used for extracting an interested region, then, the artificial design feature is used for extracting features, and finally, a classifier is used for classification and identification. However, the problems of poor effect of the region selection strategy, low robustness of manually extracted features, high complexity and the like often occur.
In recent years, with the continuous development of deep learning and convolutional neural networks, deep learning-based technologies are combined with products and used on a large scale. An important application branch in deep learning is vehicle perception, which is realized by using a vehicle detection technology, and in the prior art, an object detection algorithm based on deep learning is often used for vehicle detection. Representative vehicle detection algorithms can be divided into two broad categories, the first category being a two-stage detection model, and the representative algorithms are: R-CNN, Fast R-CNN, Faster R-CNN, RFCN, Mask-RCNN and Trident net as the basis of their algorithms and their improved networks, another category is a single-stage detection model, the representative algorithm is: a series of excellent algorithms such as the YoLO series, SSD, DSSD, DSOD, peleNet, retineNet, and shufflennet series.
At present, in the prior art scheme, a single-stage ssd (single shot multi box detector) algorithm is used to complete a vehicle detection task, and characteristics of a vehicle-road cooperative application scenario are combined: under the environment of multiple vehicles, multiple vehicle networking scenes are triggered simultaneously, the response speed of messages, the processing logic of concurrent scenes, human-vehicle interaction modes and the like are achieved, and in order to guarantee the timely updating of vehicle positions, each vehicle needs to receive and send vehicle body coordinate information at high frequency simultaneously. The established core algorithm, namely vehicle detection, is to ensure that the relevant information of the vehicle keeps high accuracy and the real-time property of information processing. The single-stage ssd (single shot multi-box detector) is a multi-target detection algorithm that directly predicts a target class and a bounding box. The SSD algorithm can achieve the same effect by synthesizing feature maps of different convolution layers. The main network structure of the algorithm is VGG16, the last two fully connected layers are changed into convolutional layers, and then 4 convolutional layers are added to construct the network structure. For output feature maps of 6 different convolution layers, the sizes of the feature maps are {38 × 38, 19 × 19, 10 × 10, 5 × 5, 3 × 3 and 1 × 1}, respectively, convolution is carried out by using two different convolution kernels of 3 × 3, and a confidence for output classification is generated, wherein each default box generates 21 classification confidences; one output regression localization, each default box generates 4 coordinate values (x, y, w, h). In addition, these 6 feature maps also generate a prior box (generated are coordinates) through the PriorBox layer. The number of default boxes per layer in the above 6 feature maps is given, and a total of 8732 prediction boxes. And finally, combining the three calculation results respectively and then transmitting the result to a loss layer, calculating loss by using a loss function and then deriving updating parameters layer by layer, if the input of the picture is scaled to 300x300, the average accuracy of the technology on a public test data set VOC2007 is 74.3%, and the forward calculation speed on a GTX titan x display card is used for reaching 46 pictures per second.
The prior art has the following disadvantages:
(1) the original SSD algorithm is used for detecting vehicles, and the requirement of accurate positioning of vehicle positions in a vehicle road cooperative application scene is difficult to meet;
(2) the vehicle detection is used as an entrance of vehicle related attribute analysis, and the subsequent attribute identification occupies considerable calculation time and needs to improve the vehicle detection speed.
In conclusion, the proposal mainly aims at the problem of vehicle detection at the roadside end of the vehicle and road cooperative application scene, and provides a roadside end vehicle sensing method based on the SSD improved algorithm. The method takes an SSD algorithm as a basic framework, improves the speed of vehicle detection by using improved feature extraction as a backbone network, and improves the accuracy of vehicle position location by using an improved target detection location loss function calculation mode. Test results show that the method has better detection effect on the cooperative application scene of the vehicle and the road compared with the original vehicle detection algorithm.
Disclosure of Invention
In order to solve the technical problem, the invention provides a roadside end vehicle sensing method based on an SSD (solid State disk) improved algorithm.
The invention is realized by the following technical scheme:
a road side end vehicle sensing method based on an SSD improved algorithm comprises the following steps:
s1: data preprocessing, namely marking and preprocessing vehicle target data through a mosaic online data enhancement algorithm based on a buffer frame SSD;
s2: training a model, namely extracting the vehicle picture features by adopting an improved feature extraction module based on a cross-stage partial connection mode and residual error connection fusion of resnet 18; improving a positioning loss function based on an SSD detection algorithm of a cafe framework to obtain a target positioning loss function fused with Complete-IOU;
s3: and (4) model reasoning, namely performing forward calculation on the algorithm model by using the model parameters output in the step S2 to obtain the target class and the coordinates of the prediction box.
Further, before step S1, the method further includes the steps of:
s0: and collecting vehicle picture data of the police and the gate, and marking all vehicles in the picture.
Further, in the step S1, the size of the image after the labeling is fixedly output through the mosaic online data enhancement algorithm based on the buffer frame SSD, a point is randomly selected in the region, the image is divided into four regions, and the four training images are scaled to the fixed size and secured in the region to form a new image.
Further, in step S2, the improved feature extraction module based on the cross-phase partial connection mode and the residual connection fusion of resnet18 is applied to a feature extraction backbone network, the input feature map is divided into two parts, part1 and part2 according to a proportion of 50%, part1 is not processed, part2 performs residual operation, and the last two parts are summarized in a concat mode.
Further, in the step S2, the target location Loss function LossIOUThe calculation formula of (a) is as follows:
Figure BDA0003522000940000041
Figure BDA0003522000940000042
Figure BDA0003522000940000051
Figure BDA0003522000940000052
where ρ () represents the Euclidean distance, bpredictRepresenting predicted bounding boxes, bgtRepresenting the actual bounding box, c the diagonal distance of the minimum bounding matrix of the prediction box and the actual box, wgtIndicates the width, h, of the actual bounding boxgtIndicates the height, w, of the actual bounding boxpredictIndicates the width of the prediction bounding box, hpredictIndicating the height of the predicted bounding box.
Also provided is a roadside end vehicle perception system based on the SSD improved algorithm, comprising:
the data preprocessing module is used for marking and preprocessing vehicle target data through a mosaic online data enhancement algorithm based on a buffer frame SSD;
the model training module comprises an improved characteristic extraction module and a target positioning loss function acquisition module based on the cross-stage partial connection mode and residual connection fusion of the resnet 18; the target positioning loss function acquisition module improves the positioning loss function based on an SSD detection algorithm of a caffe framework to obtain a Complete-IOU fused target positioning loss function;
and the model reasoning module is used for carrying out forward calculation on the algorithm model to obtain the target category and the forecast frame coordinate.
Further, still include: and the data acquisition module is used for collecting vehicle picture data of an alarm and a bayonet and labeling all vehicles in the picture.
Further, the size of the image after the mark is fixedly output through the mosaic online data enhancement algorithm based on the buffer frame SSD, one point is randomly selected in the area, the image is divided into four areas, and the four training images are scaled to the fixed size and are secured in the area to form a new image.
Further, the improved feature extraction module based on the cross-stage partial connection mode and residual connection fusion of resnet18 is applied to a feature extraction backbone network, an input feature diagram is divided into two parts, namely part1 and part2 according to a proportion of 50%, part1 does not perform any processing, part2 performs residual operation, and finally the two parts are summarized in a concat mode.
Further, the target localization Loss function LossIOUThe calculation formula of (a) is as follows:
Figure BDA0003522000940000061
Figure BDA0003522000940000062
Figure BDA0003522000940000063
Figure BDA0003522000940000064
where ρ () represents the Euclidean distance, bpredictRepresenting predicted bounding boxes, bgtRepresenting the actual bounding box, c the diagonal distance of the minimum bounding matrix of the prediction box and the actual box, wgtIndicates the width, h, of the actual bounding boxgtRepresenting the height, w, of the actual bounding boxpredictIndicates the width of the prediction bounding box, hpredictIndicating the height of the predicted bounding box.
The invention provides a roadside end vehicle sensing method based on an SSD (solid State disk) improved algorithm, which improves the accuracy and robustness of a model through a mosaic online data enhancement algorithm based on a buffer frame SSD. The improved feature extraction module based on the cross-stage partial connection mode and the residual error connection fusion of the resnet18 has stronger feature expression capability and reduces the time consumption of forward calculation. The SSD detection algorithm based on the caffe framework improves the positioning loss function, fuses the Complete-IOU positioning loss function, improves the accuracy of the detection frame, and enables the model to be converged more quickly.
Drawings
The present invention will be described in further detail with reference to the accompanying drawings;
FIG. 1 is an exemplary illustration of vehicle data labeling;
FIG. 2 is a diagram of a mosaic data enhancement example;
FIG. 3 is an improved feature extraction backbone network fabric element;
fig. 4 is a vehicle detection result output diagram;
fig. 5 is a vehicle detection result output diagram.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The specific embodiment process for the vehicle perception algorithm includes the following 3 parts:
a data preprocessing stage: and marking and preprocessing vehicle target data, and using a latest online mosaic data enhancement algorithm to improve the detection capability of the small targets of the vehicle.
A model training stage: in the vehicle picture feature extraction stage, an improved feature extraction backbone network structure is used, so that the model feature expression capability, the forward calculation speed of the model and the robustness of the model are improved; aiming at model classification prediction and vehicle positioning prediction, a Complete-IOU loss function is fused, so that the rule of selecting a bounding box is more reasonable, and the updating direction of gradient reduction of the model is optimized.
And (3) a model reasoning phase: and performing forward calculation on the algorithm model by using the model parameters output in the training stage to obtain the target category and the coordinates of the prediction frame.
In the data preparation phase:
1. and collecting the vehicle picture data of the police and the gate, and labeling all vehicles in the picture, as shown in the following figure 1.
2. The method comprises the steps of fixing the size of an output picture by using an online mosaic data enhancement algorithm based on an SSD frame, randomly selecting a point in the region, dividing the picture into four regions, scaling four training pictures into the fixed size to be secured in the region to form a new picture, enriching context information of the image, improving detection capability of a small vehicle target, enhancing robustness of a model, and generating a training picture by the algorithm as shown in FIG. 2.
In the model training phase: and a vehicle picture feature extraction stage, aiming at model classification prediction and vehicle positioning prediction.
1) And in the vehicle picture feature extraction stage, an improved feature extraction backbone network structure is used, so that the model feature expression capability, the forward calculation speed of the model and the robustness of the model are improved. The residual connecting module using the cross-stage partial connecting mode and resnet18 is applied to the feature extraction backbone network, and its block is shown in fig. 3.
As shown in fig. 3, a common residual network input/output picture is shown in fig. a; the improved cross-stage part connecting network is shown in a graph b, an input feature graph is divided into two parts, namely part1 and part2, according to the proportion of 50%, part1 does not perform any processing, part2 performs residual operation, and the two parts are summarized in a concat mode. The rest part, the whole network framework is unchanged.
2) Aiming at vehicle positioning prediction, a complex-IOU loss function is fused, so that the rule of selecting a bounding box is more reasonable, and the updating direction of gradient decline of the model is optimized. The target classification and localization prediction loss function of a conventional SSD vehicle detection algorithm is as follows:
Figure BDA0003522000940000091
Figure BDA0003522000940000092
Figure BDA0003522000940000093
wherein the target classification loss function LclsTypically, a cross entropy loss function is used; target location loss function LlocUsing smooth l1 loss function, p and u are the predicted result and true class, respectively, tuRepresents the regression results and v represents the true target box. Since the objective of regression is not bound, directly increasing the weight of regression loss makes the model more sensitive to samples with sample loss greater than 1, and is not beneficial to the training of the model, so the scheme slightly increases the position loss value for samples with sample loss less than 1 to balance the loss functions of classification loss and localization loss.
The invention improves the positioning prediction Loss function and provides a new target detection positioning Loss function LossIOUIn place of Lloc(tuV) is represented as follows:
Figure BDA0003522000940000094
Figure BDA0003522000940000095
Figure BDA0003522000940000096
Figure BDA0003522000940000097
where ρ () represents the Euclidean distance, bpredictRepresenting predicted bounding boxes, bgtRepresenting the actual bounding box, c the diagonal distance of the minimum bounding matrix of the prediction box and the actual box, wgtIndicates the width, h, of the actual bounding boxgtRepresenting the height, w, of the actual bounding boxpredictIndicates the width of the prediction bounding box, hpredictRepresenting predicted edgesThe height of the bounding box.
In the model reasoning phase: by using the model parameters output in the training phase, the algorithm model is calculated forward to obtain the target class and the coordinates of the prediction frame, and the detection results are shown in fig. 4 and 5, to sum up, it can be seen from fig. 4 and 5 that the vehicles in the image can be effectively detected, and the accuracy of the detection frame is high.
The invention takes an SSD detection algorithm as a basic frame, and a mosaic online data enhancement algorithm is newly added, so that the accuracy and the robustness of the model are improved; by taking an SSD detection algorithm as a basic framework, a feature extraction network is improved, the feature expression capability of forward calculation is improved, and the time consumption of forward calculation is reduced; by taking an SSD detection algorithm as a basic framework, a vehicle detection positioning loss function is improved, the accuracy of a detection frame is improved, and a model is converged more quickly.
The present invention also provides a computer readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of a method for roadside vehicle awareness based on an SSD improvement algorithm.
The invention also provides computer equipment which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the steps of the road side end vehicle perception method based on the SSD improved algorithm when executing the program.
The above-mentioned embodiments are provided to further explain the objects, technical solutions and advantages of the present invention in detail, and it should be understood that the above-mentioned embodiments are only examples of the present invention and are not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement and the like made without departing from the spirit and scope of the invention are also within the protection scope of the invention.

Claims (10)

1. A road side end vehicle sensing method based on an SSD improved algorithm is characterized by comprising the following steps:
s1: data preprocessing, namely marking and preprocessing vehicle target data through a mosaic online data enhancement algorithm based on a buffer frame SSD;
s2: training a model, namely extracting the vehicle picture features by adopting an improved feature extraction module based on a cross-stage partial connection mode and residual error connection fusion of resnet 18; improving a positioning loss function based on an SSD detection algorithm of a cafe framework to obtain a target positioning loss function fused with Complete-IOU;
s3: and (4) model reasoning, namely performing forward calculation on the algorithm model by using the model parameters output in the step S2 to obtain the target class and the coordinates of the prediction box.
2. The road side end vehicle sensing method based on SSD improving algorithm of claim 1, further comprising, before step S1, the steps of:
s0: and collecting vehicle picture data of the police and the gate, and marking all vehicles in the picture.
3. The method for roadside end vehicles sensing based on SSD improved algorithm of claim 2, wherein in step S1, the labeled picture size is fixed and output through the mosaic online data enhancement algorithm based on the buffer frame SSD, a point is randomly selected in the area, the picture is divided into four areas, and four training pictures are scaled to the fixed size and secured in the area to form a new picture.
4. The method for roadside end vehicle perception based on SSD improved algorithm of claim 1, wherein in step S2, the improved feature extraction module based on the cross-phase partial connection mode and residual join fusion of resnet18 is applied to the feature extraction backbone network, the input feature map is divided into two parts of part1 and part2 according to 50% ratio, part1 does not do any processing, part2 does residual operation, and the last two parts are summarized in the form of concat.
5. The road side end vehicle sensing method based on SSD improving algorithm of claim 1, wherein in step S2,the Loss function Loss of target positioningIOUThe calculation formula of (a) is as follows:
Figure FDA0003522000930000021
Figure FDA0003522000930000022
Figure FDA0003522000930000023
Figure FDA0003522000930000024
where ρ () represents the Euclidean distance, bpredictRepresenting predicted bounding boxes, bgtRepresenting the actual bounding box, c the diagonal distance of the minimum bounding matrix of the prediction box and the actual box, wgtIndicates the width, h, of the actual bounding boxgtRepresenting the height, w, of the actual bounding boxpredictIndicates the width of the prediction bounding box, hpredictIndicating the height of the predicted bounding box.
6. A roadside end vehicle perception system based on an SSD (solid State disk) improved algorithm is characterized by comprising:
the data preprocessing module is used for marking and preprocessing vehicle target data through a mosaic online data enhancement algorithm based on a buffer frame SSD;
the model training module comprises an improved characteristic extraction module and a target positioning loss function acquisition module based on cross-stage partial connection mode and residual error connection fusion of resnet 18; the target positioning loss function acquisition module improves the positioning loss function based on an SSD detection algorithm of a caffe framework to obtain a Complete-IOU fused target positioning loss function;
and the model reasoning module is used for carrying out forward calculation on the algorithm model to obtain the target category and the coordinate of the prediction box.
7. The system of road-side end vehicle perception based on SSD refinement algorithms of claim 6, further comprising: and the data acquisition module is used for collecting vehicle picture data of an alarm and a bayonet and labeling all vehicles in the picture.
8. The method as claimed in claim 7, wherein the labeled picture size is fixedly output through the spectrum online data enhancement algorithm based on the buffer frame SSD, a point is randomly selected in the region, the picture is divided into four regions, and four training pictures are scaled to the fixed size and secured in the region to form a new picture.
9. The method as claimed in claim 6, wherein the improved feature extraction module based on the cross-phase partial connection mode and the residual error connection fusion of resnet18 is applied to the feature extraction backbone network, the input feature map is divided into two parts, namely part and part2, according to a proportion of 50%, part1 does not perform any processing, part2 performs residual error operation, and the last two parts are summarized in a concat mode.
10. The method of claim 6, wherein the target location Loss function Loss is LossIOUThe calculation formula of (c) is as follows:
Figure FDA0003522000930000031
Figure FDA0003522000930000032
Figure FDA0003522000930000033
Figure FDA0003522000930000034
where ρ () represents the Euclidean distance, bpredictRepresenting predicted bounding boxes, bgtRepresenting the actual bounding box, c the diagonal distance of the minimum bounding matrix of the prediction box and the actual box, wgtIndicates the width, h, of the actual bounding boxgtRepresenting the height, w, of the actual bounding boxpredictIndicates the width of the prediction bounding box, hpredictIndicating the height of the predicted bounding box.
CN202210182687.XA 2022-02-25 2022-02-25 Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm Pending CN114639084A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210182687.XA CN114639084A (en) 2022-02-25 2022-02-25 Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210182687.XA CN114639084A (en) 2022-02-25 2022-02-25 Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm

Publications (1)

Publication Number Publication Date
CN114639084A true CN114639084A (en) 2022-06-17

Family

ID=81947411

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210182687.XA Pending CN114639084A (en) 2022-02-25 2022-02-25 Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm

Country Status (1)

Country Link
CN (1) CN114639084A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116503779A (en) * 2023-04-26 2023-07-28 中国公路工程咨询集团有限公司 Pavement casting object identification system and method

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116503779A (en) * 2023-04-26 2023-07-28 中国公路工程咨询集团有限公司 Pavement casting object identification system and method

Similar Documents

Publication Publication Date Title
CN106951830B (en) Image scene multi-object marking method based on prior condition constraint
CN110246141B (en) Vehicle image segmentation method based on joint corner pooling under complex traffic scene
CN109492583A (en) A kind of recognition methods again of the vehicle based on deep learning
Geng et al. Using deep learning in infrared images to enable human gesture recognition for autonomous vehicles
CN113379771B (en) Hierarchical human body analysis semantic segmentation method with edge constraint
CN112836657A (en) Pedestrian detection method and system based on lightweight YOLOv3
CN112801027A (en) Vehicle target detection method based on event camera
CN113850136A (en) Yolov5 and BCNN-based vehicle orientation identification method and system
CN113989744A (en) Pedestrian target detection method and system based on oversized high-resolution image
Gu et al. Embedded and real-time vehicle detection system for challenging on-road scenes
CN113378675A (en) Face recognition method for simultaneous detection and feature extraction
CN114049572A (en) Detection method for identifying small target
Shi Object detection models and research directions
Sun et al. Challenges of designing computer vision-based pedestrian detector for supporting autonomous driving
CN115019274A (en) Pavement disease identification method integrating tracking and retrieval algorithm
Li et al. Real-time tracking algorithm for aerial vehicles using improved convolutional neural network and transfer learning
CN114708566A (en) Improved YOLOv 4-based automatic driving target detection method
CN114332921A (en) Pedestrian detection method based on improved clustering algorithm for Faster R-CNN network
CN114639084A (en) Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm
CN112668662B (en) Outdoor mountain forest environment target detection method based on improved YOLOv3 network
CN112926426A (en) Ship identification method, system, equipment and storage medium based on monitoring video
CN110503049B (en) Satellite video vehicle number estimation method based on generation countermeasure network
CN117115412A (en) Small target detection method based on weighted score label distribution
CN116798118A (en) Abnormal behavior detection method based on TPH-yolov5
CN116740124A (en) Vehicle tracking and license plate recognition combined detection method based on improved YOLOv8

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination