Millimetre-wave (mmWave) radar has emerged as an attractive and cost-effective alternative for human activity sensing compared to traditional camera-based systems. mmWave radars are also non-intrusive, providing better protection for user privacy. However, as a Radio Frequency based technology, mmWave radars rely on capturing reflected signals from objects, making them more prone to noise compared to cameras. This raises an intriguing question for the deep learning community: Can we develop more effective point set-based deep learning methods for such attractive sensors? To answer this question, our work, termed MiliPoint, delves into this idea by providing a large-scale, open dataset for the community to explore how mmWave radars can be utilised for human activity recognition. Moreover, MiliPoint stands out as it is larger in size than existing datasets, has more diverse human actions represented, and encompasses all three key tasks in human activity recognition. We have also established a range of point-based deep neural networks such as DGCNN, PointNet++ and PointTransformer, on MiliPoint, which can serve to set the ground baseline for further development.

In this paper, we describe a method we have employed at the University of Bristol to improve the undergraduate project experience. We describe the methodology we employ, which consists of a pre-built environment, close supervision by a senior student, and compiled reference material. We then compare the methodology with its absence using the frequency of undergraduates publishing as our metric. We observe a noticeable increase in the number of undergraduate publications under the new methodology, as well as a number of unexpected benefits for the students.

Various 3D reconstruction methods have enabled civil engineers to detect damage on a road surface. To achieve the millimetre accuracy required for road condition assessment, a disparity map with subpixel resolution needs to be used. However, none of the existing stereo matching algorithms are specially suitable for the reconstruction of the road surface. Hence in this paper, we propose a novel dense subpixel disparity estimation algorithm with high computational efficiency and robustness. This is achieved by first transforming the perspective view of the target frame into the reference view, which not only increases the accuracy of the block matching for the road surface but also improves the processing speed. The disparities are then estimated iteratively using our previously published algorithm where the search range is propagated from three estimated neighbouring disparities. Since the search range is obtained from the previous iteration, errors may occur when the propagated search range is not sufficient. Therefore, a correlation maxima verification is performed to rectify this issue, and the subpixel resolution is achieved by conducting a parabola interpolation enhancement. Furthermore, a novel disparity global refinement approach developed from the Markov Random Fields and Fast Bilateral Stereo is introduced to further improve the accuracy of the estimated disparity map, where disparities are updated iteratively by minimising the energy function that is related to their interpolated correlation polynomials. The algorithm is implemented in C language with a near real-time performance. The experimental results illustrate that the absolute error of the reconstruction varies from 0.1 mm to 3 mm.

The detection of multiple curved lane markings is still a challenge for advanced driver assistance systems today, due to interference such as road markings and shadows cast by roadside structures and vehicles. The vanishing point Vpcontains the global information of the road image. Hence, Vp-based lane detection algorithms are quite insensitive to interference. When curved lanes are assumed, Vpshifts with respect to the rows of the image. In this paper, a Vpfor each individual row of the image is estimated by first extracting a Vpy(vertical position of the Vp) for each individual row of the image from the v-disparity. Then, based on the estimated Vpy's, a 2-D Vpx(horizontal position of the Vp) accumulator is efficiently formed. Thus, by globally optimizing this 2-D Vpxaccumulator, globally optimum Vps for the road image are extracted. Then, estimated Vps are utilized for multiple curved lane marking detection on nonflat road surfaces. The resultant system achieves a detection rate of 99% in 1862 frames of six stereo vision test sequences.

In this paper, we introduce a robust lane detection and tracking algorithm to cope with complex scenarios and to decrease the effect of thresholds. For lane feature extraction, an extension to the symmetrical local threshold (SLT) is proposed to improve the feature map and obtain orientation information. Then, while creating a Hough accumulator, obtained orientation information is used to decrease computational complexity (≈ 60 times) and acquire a clearer accumulator. The left and right lanes are categorized by applying a mask on the Hough accumulator, which leads to low computational complexity and reduced sensitivity to thresholding. To quantify the new feature map, we used ground truth lane markings from the RoMa Datasets and the optimum true positive (TP) to positive (P) ratio increased from 69% to 86% on average, compared to the SLT. The successful lane detection rate calculated from more than 10K frames is, 96.2%, demonstrating the robustness of the system.

In this paper, a stereo vision based pothole detection system is proposed. Using the disparity map generated from an efficient disparity calculation algorithm, potholes can be detected by their distance from the fitted quadratic road surface. The system produces the size, volume and position of the potholes which allows the pothole repair to be prioritised according to its severity. The quadratic road surface model allows for camera orientation variation, road drainage and up/down hill gradients. Experimental results show robust detection in various scenarios.

Recently many cellular and WLAN standards are considering operation in the TV bands due to enhanced coverage and access to additional bandwidth. To accurately assess the performance of wireless communication systems operating in the TeleVision White Space (TVWS) spectrum, a new channel model is required that accounts for the unique propagation characteristics in the UHF band. This paper proposes an extension to the existing 3GPP/ITU channel model for operation at frequencies from 400 MHz (bottom of the TVWS band) up to 5 GHz (top of the WLAN band). The channel model is suitable for urban environments and the raw propagation data was generated using a state-of-the-art 3D ray-tracing tool. Results are presented as cumulative distribution functions of total received power, RMS delay spread, K-factor and RMS angular spread (azimuth and elevation). The probability of line-of-sight as function of distance and best-fit parameters for large and small scale fading are also presented.

To maintain a high quality of service to the increasing number of smartphone users, additional spectrum is required. The TV white space bands provide a good opportunity to achieve this goal. They can be used as standalone spectrum or aggregated with other licensed bands to increase the total available bandwidth. This paper compares the downlink throughput performance of LTE-Advanced in two different frequency bands. It also addresses the impact of smartphone orientation, with results quoted for three different elevation angles. We consider the higher LTE-Advanced band at 2.6 GHz and the TV white space band at 800 MHz. The radio channel is modelled using a state-of-the-art 3D ray-tracing tool combined with measured 3D radiation patterns for the base station and handset antennas. The throughput performance for a large number of base station and mobile terminal locations is investigated in two different UK cities. We use the computationally efficient received bit information rate algorithm to compute packet error rate as a function of channel structure and SNR. The approach reduces simulation run time by a factor of more than 300. Similar average throughput vs SNR results are observed in both frequency bands. However, higher K-factor and total received power levels are observed when the user equipment is tilted to 45oin elevation. Throughput results show that the efficiency of carrier aggregation between LTE and TVWS bands depends on the cell size and the type of urban environment.

This paper presents a local disparity calculation algorithm on calibrated stereo images based on cost aggregation. Unlike most of the existing cost aggregation methods which are mainly based on the grouping of colour similarities, the proposed algorithm is grouped by local cost similarities. The proposed algorithm also applies a bilateral filter to enhance the normalised cost volume and, then, uses the winner-take-all technique to select the correspondence candidates. Finally, a quadratic polynomial interpolation is performed using the candidates and their neighbourhood values to achieve sub-pixel disparity resolution. The experimental results indicate that the proposed algorithm is able to provide dense disparity maps with sub-pixel resolution and achieves better accuracy compared to two similar stereo matching algorithms.

In this paper the authors are addressing the concerns associated with fast growing DSP chips and tools and the impact they have on teaching DSP implementation. The authors also provide solutions, advice and suggestions on how to select a DSP, set a DSP implementation course and the associated laboratory hardware and software that fit a specific application. To help differentiate between the multitude of high performance DSPs, vendors are developing market-specific DSPs as an alternative to more generic DSPs. This is putting pressure on educators to update courses and reinvest in hardware and software more often than for traditional courses. For the less experienced or new educators, it is a daunting task to cope with this fast evolving technology, to make an informed choice and select the right tools. Choosing the wrong tools not only can be very expensive and time consuming but can also have a negative impact on teaching and learning outcomes.

This paper presents an obstacle detection system which is robust to non-flat road surface and interference of illumination. A 3D camera is used to generate depth information without the need of camera calibration. The depth map is then transformed into U-V-disparity domain, where obstacles and ground surface are projected as lines. Hough Transform is employed to extract line features; it has been modified to fit the characteristic of the U-V-disparity in order to boost the speed and accuracy. In addition, steerable filters are applied to the u-v histogram before Hough Transform for noise reduction. By categorising extracted lines according to their position and posture, road surface and on-road obstacles can be detected. Finally, results obtained using both U and V disparity maps are combined to eliminate road side surface and post processing. Experiments show that the proposed system is Able to detect obstacles accurately under various environments.