Table 2.
Record of the general characteristics of the selected studies.
| Author (s), year of publication, country | Objectives | Methods and sample | Results (main outcomes) | Key limitations |
|---|---|---|---|---|
| Brijs et al., 2022 (Belgium) (47) | Design and evaluation of a driver assistance system for overtaking cyclists. | A sample of 48 drivers performed the established route with the system activated and deactivated. | The system had an impact on the duration of the overtaking phase, the lateral clearance in the overtaking phase and the hazard time in the process of the overtaking maneuver. | Repeated use of the system may reduce its effectiveness or cause learning effects that influence the results. |
| Siebke et al., 2023 (Germany) (48) | Emergency braking system (AEB) for cyclist detection in urban intersection scenarios is evaluated. | Driving simulations are used to evaluate 240,000 road situations with various sensor opening angles. | The presented approach allows examining the entire scenario space randomly, which minimizes the potential loss of information in risky situations. | No limitations are apparent |
| Rasch and Dozza, 2020 (Sweden) (49) | Design of a control model based on logistic regression to avoid false-positive ADAS activations. | Data from an experiment on a test track are recorded to establish the model. | It manifests limited ability to predict the probability and confidence of drivers braking and turning while approaching a cyclist during an overtaking, thus improving ADAS. | Small sample size |
| Kovaceva et al., 2022 (Sweden) (50) | The potential impact of forward collision warning systems on cyclist protection in overtaking situations is evaluated. | Drivers’ reactions to the warning were analyzed, combining the data with accident frequency and an injury risk model. | With the driver response model applied to the warning system, cyclist fatalities were reduced by 53 to 96% and serious injuries by 43 to 94%. | Simulations did not include responses other than braking (e.g., turning). |
| Schindler and Piccinini, 2021 (Sweden) (51) | The response of drivers in two conflict scenarios with vulnerable road users is analyzed. | A group of 13 people took a tour driving a truck equipped for data recording. | Drivers adapted their kinematic and visual behavior in situations where vulnerable road users were crossing the intersection, compared to the baseline route. | Small sample size |
| Char and Serre, 2020 (France) (52) | Analyzed accidents between cars and cyclists to determine potential improvements for vehicle active safety systems. | Analysis of 2,261 accidents involving cars and cyclists. Safety systems are applied in the most likely incident scenarios. | A field of view of 60° and a range of 35 m would allow detection of most cyclists in accident scenarios. With a 60° field of view, 51% of cyclists could be detected up to 4 s before the crash and 72% up to 1 s before. | The sample is not representative of the national proportion of accidents. |
| Limani et al., 2022 (Belgium) (53) | PowerCam, a system that enables compatibility between 802.11p and conventional Wi-Fi networks, is presented to connect cars with the cell phones of vulnerable users. | A standard wifi AP is included in the roadside unit’s broadcast radio and another just outside it to verify that messages are broadcast. | This methodology enables low bit-rate communication between devices without requiring formal association or authentication. The results demonstrate the system’s ability to deliver messages in a timely manner to users. | May have a latency of 2 s, not being effective in specific scenarios. |
| Cara et al., 2015 (Netherlands) (54) | A scenario classification algorithm using machine learning is proposed to evaluate ADAS systems for cyclist protection. | A data set consisting of 99 realistic cycling scenarios recorded by a vehicle equipped with instrumentation is obtained. | An accuracy of 87.9% is achieved in the classification of the data obtained, and the execution time of 45.8 microseconds supports the suitability of the algorithm for fine online applications. | No limitations are apparent |
| Puller et al., 2023 (Germany) (55) | A V2X-based turning aid designed to mitigate collisions with vulnerable participants in traffic is presented. | Generate information for advanced driver assistance systems to use, even when the sensors do not detect the object in the foreground, and increase awareness of crossings. | The application faces challenges in terms of user acceptance, so a key challenge for ADAS functions is to maintain a low false positive rate so that users do not lose confidence in its accuracy. | No limitations are apparent |
| Brijs et al., 2021 (Belgium) (56) | The impact of an advanced driver assistance system for cyclist overtaking is analyzed. | A driving simulator is used for the experiment in which there are three phases of warning priority: normal accident, hazard, and avoidable | A positive effect on lateral clearance was observed with ADAS presence, familiarity with the system, driving experience, and experience as a cyclist. A negative effect of cyclist maneuvering from the edge of the lane to the center of the lane, cyclists riding parallel, driver age, and self-reported aggressive driving. | No apparent limitations |
| Kovaceva et al., 2019 (Sweden) (57) | The combination of factors affecting the limits of drivers’ comfort zone when overtaking cyclists in a naturalistic environment is analyzed. | Naturalistic driving data from UDRIVE, a European naturalistic driving study, is analyzed. | The higher the speed of the car, the higher the driver’s comfort zone limits when approaching and passing, but the presence of an approaching vehicle decreases it when overtaking. | Limited generalization of the data set |
| Kovaceva et al., 2020 (Germany) (58) | The safety benefit of autonomous steering and emergency braking systems for the protection of cyclists and pedestrians is evaluated | Data from a simulation based on data from the German In-Depth Accident Study (GIDAS-PCM) were combined with real-world test results. | A systematic way of combining results from different sources is indicated, showing the positive effects of the evaluated system. | Other scenarios may require the application of an extension of the current model. |
| Anaya et al., 2015 (Spain) (59) | The effectiveness of V2X systems in detecting vulnerable road users is evaluated. | Two tests are performed to test the correct detection by the vehicle of both motorists and cyclists. | In both tests the vehicle correctly detects the vulnerable user even in blind spots when the distance between the two vehicles is less than 30 meters. | No limitations are evident |
| Guerrieri and Parla, 2021 (Italy) (60) | The aim is to obtain a program capable of detecting vulnerable road users by calculating their distance and speed, in order to be able to act from the streetcar. | Images obtained along the route of a streetcar are analyzed and processed by neural networks to obtain different parameters. | The system is able to correctly estimate the approach speed of pedestrians, cyclists and other vehicles. | No limitations are evident |
| Azadani and Boukerche, 2021 (Canada) (61) | The aim is to obtain the position of cyclists in motion in order to improve the detection capability of the ADAS. | Two different real scenarios are simulated in which the cyclist is in a position not visible from the vehicle, calculating his position. | In both scenarios, the ultrasonic sensors installed in different cars were able to locate the cyclist’s position and share the information among several vehicles to keep the cyclist located. | No limitations shown |
| Chen et al., 2018 (China) (62) | The improvement of pedestrian and cyclist identification by unifying 3 different detection methodologies is evaluated. | Evaluations of the proposed detection methodology are carried out by comparing it with other detection methods. | The proposed method shows a higher efficiency in recognizing pedestrians and cyclists than other methods used. | No apparent limitations |
| Ucińska and Pełka, 2021 (Poland) (63) | To analyze the effectiveness of the automatic braking system (AEB) in different situations in front of VRU. | 44 trials of 4 tests are performed in which different scenarios are presented analyzing the AEB activity. | The different tests show the difficulty of the AEB system both in braking in time and in recognizing properly the VRUs present on the road. | Low sample size |
| Xu et al., 2020 (China) (64) | A neural network is trained to improve the differentiation between cyclists and motorists, as well as co-detection. | Comparison of the results obtained between the proposed method and the previously existing ones on several databases on the Internet consisting of more than one million images (4,300 of cyclists and motorists). | The results show a higher accuracy in differentiating between cyclists and motorists by up to 30%. | No apparent limitations |
| Duan et al., 2017 (China) (65) | Analyzing the braking behavior of drivers to improve the performance of ADAS braking systems on bicycles. | 3 types of scenarios were simulated from the accident data collected and 25 drivers were tested. | The application of the data obtained can help to improve bicycle AEB systems and thus reduce the number of accidents. | Small sample size |
| Char et al., 2022 (France) (66) | Potential effects of earlier brake activation by drivers with a collision warning system are quantified in simulated car-to-cyclist accident scenarios. | A parametric analysis is performed by varying the field of view of the detection sensor, the activation time of the forward collision warning system and the reaction delay time of the driver to the forward collision warning system. | A 70° field of view, a system activation time of 2.6 s before impact and a driver reaction of 0.6 s to the warning system has a positive outcome in 82% of accident cases, with 78% avoiding and 4% mitigating the crash. | No apparent limitations |