In recent years there has been an increase in the number of research and developments in deep learning solutions for object detection applied to driverless vehicles. This application benefited from the growing trend felt in innovative perception solutions, such as LiDAR sensors. Currently, this is the preferred device to accomplish those tasks in autonomous vehicles. There is a broad variety of research works on models based on point clouds, standing out for being efficient and robust in their intended tasks, but they are also characterized by requiring point cloud processing times greater than the minimum required, given the risky nature of the application. This research work aims to provide a design and implementation of a hardware IP optimized for computing convolutions, rectified linear unit (ReLU), padding, and max pooling. This engine was designed to enable the configuration of features such as varying the size of the feature map, filter size, stride, number of inputs, number of filters, and the number of hardware resources required for a specific convolution. Performance results show that by resorting to parallelism and quantization approach, the proposed solution could reduce the amount of logical FPGA resources by 40 to 50%, enhancing the processing time by 50% while maintaining the deep learning operation accuracy.

Due to a point cloud’s sparse nature, a sparse convolution block design is necessary to deal with its particularities. Mechanisms adopted in computer vision have recently explored the advantages of data processing in more energy-efficient hardware, such as the FPGA, as a response to the need to run these algorithms on resource-constrained edge devices. However, implementing it in hardware has not been properly explored, resulting in a small number of studies aimed at analyzing the potential of sparse convolutions and their efficiency on resource-constrained hardware platforms. This article presents the design of a customizable hardware block for the voting convolution. We carried out an in-depth analysis to determine under which conditions the use of the voting scheme is justified instead of dense convolutions. The proposed hardware design achieves an energy consumption about 8.7 times lower than similar works in the literature by ignoring unnecessary arithmetic operations with null weights and leveraging data dependency. Access to data memory was also reduced to the minimum necessary, leading to improvements of around 55% in processing time. To evaluate both the performance and applicability of the proposed solution, the voting convolution was integrated into the well-known PointPillars model, where it achieves improvements between 23.05% and 80.44% without a significant effect on detection performance.